# AI AGENTS MASTERCLASS 4 HOURS: Build & Sell (2026)

Channel: Michele Torti
Video: https://www.youtube.com/watch?v=0C6xYvQhjwQ
Duration: 3 hr 42 min
Language: English
Words: 48071
Transcript page: https://viewrankai.com/tools/youtube-transcript/0C6xYvQhjwQ

---

[0:00] Welcome to the full course on how to build and sell AI agents to businesses. By the end, you'll know exactly what AI agents are and how they work. You will have watched me build five different agents on four different platforms. You will know how to find a business that pays you for them, how to run the call that closes them, and how to price and package it in a way where the business will say yes. And I'm not going to sell you a dream. I'm not going to tell you that you can make 10K in 7 days because

[0:24] I personally cannot promise you that. And the people who do are setting you up to fail. What I can tell you is my own experience going from studying to become an investment banker, setting myself up for a 9 to-ive career that honestly I didn't want to actually going all in into AI in 2024 and scaling my own AI agency to six figures in 14 months having worked with over 50 businesses from small shops to enterprise clients and teaching thousands on how they can do the exact same thing. So everything in here is what actually works with real clients and real [music] money, not just what I think sounds good. And one honest

[0:57] warning before we start, this is for people who are genuinely committed and serious about building a real skill and building a real business, not for people who are either wanting the button that prints money or wanting to just learn AI for fun. Cuz if that's you, there are tons of videos that are much better for that. But if you're serious about making real money with AI, then you are in the right place. Now, I've also left chapter markers in the description below in case you want to continue where [music] you

[1:19] left off and come back to the video. Otherwise, start here because each part builds on the next. With that being said, guys, let's dive in. All right, so AI agents from the ground up. What they really are, how they think, how they take action in the real world, how we can use them to make money, but most importantly, why should we even learn these things? So, in terms of the opportunity itself, the market of AI agents, it's growing around 45% every single year, right? Meaning that you guys, especially at the time of this

[1:48] recording, you are extremely early. There's going to be a window of three to five years where businesses are going to have to adapt to these AI agents or they will get left behind. I know it's something that people say, but it's the truth because every single business will need to be AI first, AI native because of these technologies that are coming out which are way better than having any employees because they work faster, they don't take holidays, they don't complain, and they work 24/7. Now, as of 2025, the market size of AI agents was 7.8 billion. As of 2030 projecting, it's 50 billion. My assumption is that this

[2:21] is probably going to double or even triple as well because projections only they're pretty inaccurate at the time of the the actual production especially with AI but to the point here it's growing 45% year on year right hence why you see most of it in North America so the US and Canada you have Asia you have Europe and the rest of the world as well right all right so now the question becomes what actually is an AI agent like what does it mean the word is everywhere but maybe you might not really know the definition of tactically what it means so An AI agent is a

[2:50] software that can think for itself. It makes decisions and takes action to reach a specific outcome, right? Without you being there every step of the way. Really, really important. Without you being there every step of the way, so it can reason about the situation, which is the thinking part, it can decide as to

[3:08] what to do next like a normal person. And it can act, which means things in the real world. And so the reason why these AI agents are taking over the world in my opinion is because they are a form of a human employee in this case AI employees. The only difference is that they do it 10 times faster 24/7 without taking any breaks. Now of course there's the argument that these are not going to replace every single human in the world because you still need the creativity sometimes you still need the human thought but for the most part for most of the work that's in the economy

[3:39] this can be automated using AI agents. [music] Now with this said there's a full transition that happened over time because we didn't always start with AI agents. This is something that recently came in 2025 2026 and beyond right and so what actually happened like how did we go from where we were to where we are. So up until 2010 we had the era of the manual code. So this is software engineers coding every step of the way by hand right line by line. And that's where there was no automation tools there was no AI agents there was no AI but it was just manual coding know up

[4:07] until 2024 which is workflow automation tools. And by the way, we still use workflow automation tools. We will use them today when we build things. And these are tools like Zapier, Make.com, N10, which essentially take softwares and wire them together, put them together. The only difference from manual code to work with automation is that you don't need any code experience, meaning that it's no code platforms. So without coding, you're now able to do things that you could have done here in a manual code, but do it in a software that's really easy to to look at. You just have to drag and drop modules and it does everything there, right? This

[4:38] diagram shows pretty much what it looks like. It's just one box connected to another box connected to another box and so on, right? And so you start piecing things together, but again, you still have to create the automation manually. It's not really automated. You can't just say, "Hey, I want this." And it does everything for you. Then we go to this new era, the agentic era, the AI agents era. And this is where softwares can think and act for itself. So this is where you guys are here right now, which is by far again one of the biggest opportunities in the software industry that's happened in a very, very long

[5:07] time. And the reason that is is because now we don't have to code manually. We don't even have to drag and drop different automations, but we can just say, "Hey, I want X, Y, and Z done, and it has everything that it needs." And it just builds it for you. And so this is the ultimate level of like a person who has no clue how to code, no clue how to build automations, but knows plain English and knows what they want and

[5:28] just saying, "Hey, I want X, Y, and Z." And it's done. Right? Of course, there's a lot of things that go behind it, which we'll talk about today. But like the main premise of this is that now we've come to a point where like the real skill that someone needs to acquire going into AI is most times the prompting side which is how to talk to AI agents to get the thing that you want to get the outcome that you want for a business, right? Not so much the ability to understand everything although understanding the foundations is extremely important as well. Now with this said there are five different parts

[5:57] of an AI agent, right? And these five parts are basically what constitute any AI agent that you think of and see online. Every single AI agent, right? We have different tools, different software that we use to build AI agents. But fundamentally, like the foundations, they're all the same. The first part of an AI agent, which is one of the most important, is the brain, which is the

[6:15] LLM. So this is the reasoning engine. It's like the the chip inside your brain, right? The better the chip, the better it can do the actual thing, right? Assuming that the prompt is the same, the tools are the same, the instructions, all that stuff is the same. The brain is the way or the actual thing in this case the LM that uses to be able to think and this is where we can get into the discussion of whether chat GBT is better than Claude, whether Claude is better than Gemini, whether

[6:40] Gemini is better than Grock, right? There's all these different LLMs. The good thing about this is that as you guys might see in the industry right now and over time is that there are new models coming out every single week. But it is better for us because when we're building AI agents, the better the model usually the better the outcome that we get because it can think better. Right? The reasoning engine is like it's a difference between a Toyota which is Toyota engine which was chatbt two years ago and a Ferrari engine which is clawed model today. It only keeps getting better. And so with this said

[7:10] with new models coming out it's new reasoning engines. And so we're able to access LLMs that are much better over time that are able to take on more information, able to do more complex things at a faster, you know, speed and also more consistently that will enable the AI agent to give us a much better output. Now, the second part of an AI agent is the instructions. So this is the job description, the prompt telling it how to behave. Because of the fact that now we don't really have to know exactly how to build automations like step by step by plugging them in. The main skill, the main premise of this is

[7:41] that we have to know how to speak to AI agents. The same way that if you hire a human being, you have to know how to instruct a human being. If you just speak in Morse code to a human being and he won't really understand, so he doesn't really give you the output that you wanted to. And so, usually the input dictates the output. Meaning that if you put in, if you're prompted if your input is then you will get out, meaning that your output will be It doesn't really like align with what you want. And so a lot of the times where a lot of people say, "Hey,

[8:07] um, the AI agent doesn't work or it's not giving me the output that I wanted to." It usually comes down to your instructions because how many times have you been instructed to do something and you did it wrong, but it wasn't because of you. It was because your instructions were wrong and so you did what you were told to do. And so we're going to dive this in um a bit deeper as well when we look at prompting. But there are different prompting techniques that we

[8:26] do use when we give AI a specific task. It's not just as easy as saying, "I want X, Y, and Z." there's more things that come into the prompt instructions marked on formatting that I mean the format of the prompt the actual rules the the actual structure and so we'll look at that as well but the instructions here will be the input that we give to the AI agent for it to give us the output now the third part of an AI agent is what we call memory now this is uh the fact that AI agents can remember the conversation and what happened before the way that I

[8:53] personally think about memory when it comes to AI is that it has short-term memory and it has long-term memory short-term memory is the memory that it has from the conversation that you're having it right now. For example, if me and you are having a conversation right now, you will remember the conversation as we're having it. But when you have a new conversation, you will forget that we ever had the conversation. On the

[9:11] other hand, we have long-term memory. So, this is maybe details and information about a specific thing that you store in the back of your head that you can always come back to and retrieve whenever you need it. In this case, AI agents can store long-term memory using like a Google sheet or air table or whatever, some sort of software that stores information that the AI agent can always pull information from. and it has short-term memory which is a memory of the conversation that it's having right now. That is the simplest way of me explaining memory when it comes to like these AI agents specifically. The fourth

[9:38] part of an AI agent is what we call the knowledge base. So the knowledge base is the long-term memory. Meaning that it's documents and data that the AI agents can always retrieve information from to answer your question. Now the most simple example to put this is [music] a customer support agent. So I'm sure you guys have spoken to like a customer support agent before. You've asked it, hey, what's the return policy on my order? And then it gives you the answer, right? The answer comes from a document that it's, you know, trained on that is able to use to answer your question. And so this can be documents, this can be

[10:10] Google sheets, this can be uh, I don't know, images, videos, whatever it is. But it's some sort of like piece of data that it can always look at, reference whenever it has to answer your question. Then obviously the documents can be stored in a very specific way. We have vector databases, we have rag databases. It's a whole thing there, right? But the main thing to understand is that these documents are something that the AI

[10:28] agent uses to then answer your question. And by the way, we're going to use these when we're going to build. So you'll see them in action. And the last part of an AI agent and arguably the most important one in my opinion is what we call tools. I put a star here because of the importance of the actual part of the AI agent. Now the brain is important. The instructions extremely important because

[10:45] it's instructions. The memory is key. Knowledge base as well. But the tools is how it actually does everything that you want it to. It's how it updates the CRM. It's how it sends the email. It's how it books the appointment. It's how it does pretty much everything. And then everything around this is, you know, basically setting up the foundations so

[11:01] that it's able to use these effectively. It's able to use the tools without messing up, right? But the tools that it has are the ones that are able to then do all the actions that you want it to. When you hear of voice agents, when you hear of lead generation agents or customer support agents, these agents have access to these tools that are able to then access the internet or different softwares to take actions for you. So you don't have to. And so with this said, let's say that a customer asks the AI agent, where is my order? That's the first message. So where's my order? The

[11:31] first thing that the AI agent does is that it uses the brain to understand what the person is talking about and understands that they're asking about an order status. So this is where the LLM comes in the first part of the AI. Then it uses the instructions because we instruct the agent to always check the order before replying. And so it's going to use that instruction to then go to the memory which recalls the order number that they gave earlier in the conversation. Then it can use for example the knowledge base maybe a document which checks the shipping policy for delivery times. And finally it uses the tools to call the store API

[12:02] which means you know the actual store software that you use. It gets the status of the order which is this and it replies to the agent itself. And so this step-by-step instructions that you see here of how the AI agent works is what we call logs. So logs are essentially like whenever I give the AI agent a message like hey schedule a calendar event we can see exactly what steps it takes towards getting us to the outcome and so it thinks it looks at the instructions it recalls a memory it uses the knowledge base and then finally it uses the tools to then be able to do the

[12:31] actual thing itself. And so these four things are setting us up to make sure that when it does use the tools, it doesn't [music] mess up. But the tools itself are the thing that we use to actually get the information that we want. Now, with this said, how is a chatbot any different from an AI agent? Cuz I'm sure you guys have used the chat

[12:46] bots online or maybe even phone bots. They say, "Hey, press one for this. Press two for this. Press three for this." The difference is that whilst a chatbot only can talk back to you, it can only answer questions. It waits for your next message and it can't do anything in the real world. An AI agent can think and then act. It uses tools to get things done. It can make its own decisions and it completes the whole job for you. So while it's a chatbot tells

[13:09] you what to do, an AI agent does it. Hence why it's a chatbot you can just chat to it, [music] an AI agent, it's an agent that does things for you that's powered by AI. Now, the main difference between a workflow and an AI agent is that a workflow specifically, all it is, it's just a fixed series of steps that runs the same every single time. [music] So, you might have heard the term

[13:30] deterministic versus non-deterministic. What deterministic means is that you can determine, you can predict what's going to happen. And with a workflow, let's say we have a workflow that starts with new call booked, then it goes to add the lead to the CRM, then it goes to send a text to our team, and then maybe sends a thank you email, right? This right here wouldn't change as in every single time that you run it, it will always run from this to this, right? And so this is the workflow aspect of it. Meaning that it's

[13:54] rulebased and it's fully predictable. It's deterministic. It doesn't change. It doesn't go above the Slack message or above the email or above pipe drive, right? Which is the CRM. It always happens step by step based on the workflow and the rules that you put into the actual system. An AI agent on the other hand is not locked into fixed steps. is not step one, step two, step three, step four because we can have the same starting point which can lead to a different action because the agent decides what to do. So let's say we talk to an AI agent on Telegram which is a a messaging platform and we say hey make

[14:24] an email or send an email or maybe even schedule a calendar event right now. The AI agent, which in this case powered by Claude, will think through what the instructions are because it has instructions and then it will either make a calendar event or it will send an email based on the tools that it has access to. And so we can have different kinds of inputs leading to different kinds of outputs or the same input which can lead to different kinds of outputs because the AI agent is the one responsible for thinking through exactly what to do at any given point in the system itself. Hence why I say that like

[14:57] it's very good that it does that because it can think on its own. But it's also not that good if your instructions are not that good. If you say, "Hey, just think about it and do whatever," the output wouldn't really be consistent. But if you give it fixed steps like step one, step two, step three, it can use those as guard rails and then make its own decisions based on what it is that you want. Now, with the fact that these tools right here, calendar and email and CRM and all that stuff is really the thing that actually makes the AI agent so powerful because they're able to

[15:23] access all these softwares at once. The question becomes how is an AI agent able to access our softwares for us? How is it able to tap into a CRM? How is it able to send an email, schedule a calendar event if we are not doing it ourselves, right? How is it taking action for us? It's one answer and the answer is API. So API stands for application programming interface. It's essentially the process of a software speaking to a software. an AI agent speaking to the CRM, an AI agent sending an email, an AI agent updating some sort of like Slack message or Google sheet,

[15:54] right? Now, think of this as a waiter. The waiter has a chef, and it has the customers. Between the chef and the customers, there's a series of steps that it has to do. There's taking the order, there's bringing the order from the kitchen, delivering the order, making the order, taking the order, bringing the order, delivering the

[16:08] order, make the order, take the order. So, it goes on and on and on. It's a whole repetitive loop that it goes through, which is exactly what APIs are about. So instead of the customers, think of the customer as the end user, which in this case you on the other side of the screen talking to an AI agent, right? And you give a request, meaning that, hey, I want X, Y, and Z done. And so the AI agent will then use an API to then call Google Sheets, which is in this case the chef, or call Gmail, right? And so the API is the waiter that's in between you wanting something

[16:38] and something getting done. And between you and the software, it's a series of request and response. Request and response. I want this done. The response is it's done. I want to update a CRM with a new conduct. The response is the CRM has been updated. Right? And so really the main thing to understand is

[16:54] that API is how everything gets done. Now you might be asking yourself why is it that I have to learn about APIs when the software take care of that themselves and I don't have to worry about it. Well the reason that is is because when something goes wrong with claw code or with any 10 or whatever it is then you in a much better position to fix and understand what went wrong. If you understand the foundations of how everything works in the first place because then you understand what went wrong and how to fix it itself. Instead, if you're someone who doesn't really understand the foundations at all, then

[17:23] it will be much harder for you because you have no context as to how things work. All right, so APIs again are the reason to why softwares are able to speak to softwares without you having to do it yourself. But what can they actually do? Well, there's four things that they can do. Two are the ones that you're most likely going to use 90% of the time. And these two are what we call post and get. So post and get are different types of requests which are the two that you'll most likely use. So the first one is post. Post is any type of record that you have to create is you

[17:51] sending information from where you are to something. Let's say you want to send an email. That's a post request. So you'd be sending a post request to the server saying hey guys I want to send an email post. Let's say you want to add a contact to a CRM. Post request. Hey server I'm sending a post request to add a contact to my CRM. Sending information

[18:07] from where you are to something else. Right? outbound message or maybe even sending a message on LinkedIn. So whenever you send a message on LinkedIn in the back end, you're actually sending a post request to LinkedIn saying hey I want to send a message to someone and that will be a post request. Now on the other hand a get request is a complete opposite. So while post is sending get is getting right you're getting information from any record from the software. So let's say you are using Gmail and you want to retrieve the past five emails of today. So you'd be saying, "Hey server, hey Gmail server,

[18:38] hello. How are you? Um, I want to get something. I want to get the past uh five emails of today using a get request." And so it will give you the the actual thing itself. Or I don't know like your CRM. You want to pull the last 10 contacts who opted in to your Facebook ads. Hey server, I'm sending a get request to get my last 10 contacts

[18:58] who have opted in to my Facebook ads. That's a get request. And so post and get are the main ones that we're going to use. Now, this is important because I'm going to be talking about this when we actually go to the builds and I'll talk about post and get, when we have to schedule calendar events, when we have to get emails. It's uh it's going to be awesome. So, keep this in mind when we go to the next steps. Now, the last two are going to be uh put, patch, and delete. Put and patch is when you update a record. So, let's say John's name is

[19:21] not John anymore, it's James. Then you have to update the record using a put or patch request saying, "Hey server, I'm giving you a put a patch request to update a name. Can you action it? Can you do it?" And it will do it. And then we have delete. Pretty straightforward. If you want to delete any record from a software, so let's say you have a Google sheet record or some sort of item and you want to delete it, you say, "Hey server, I'm sending you a delete request to delete the line item in the Google sheet." And it doesn't. So, and so these are the four, but these two are the main

[19:48] ones that we're going to use. Again, we do use put patch and delete sometimes depending on how complex you go or what sort of functionalities you want to do, but mainly you're going to be getting information and sending information. Now there's a fundamental concept which is the fact that like an AI agent is only able to access a software if they have an API meaning that if the software is available to say hey you can send information you can retrie information you can do whatever you want with my software then it's possible but there are times where like the software itself is may you know maybe old or maybe

[20:17] outdated where it doesn't have an API and so how do we find that out and so that's where we go into API documentation so what this means is like a documentation like basically a Google doc telling us exactly what we can automate and cannot automate in a specific software. So let's say if we take Google Sheets, let's say we can only create a new row, but we cannot retrieve a row. That's where we go to the API documentation and we'll find that and then we'll see, okay, what things can we do? What things can't we do? And it will tell us. Now, how do you find the API documentation? Well, you

[20:42] can just search up the software name. So, HubSpot API documentation, ClickUp API documentation, Google Sheets API documentation. Right? Once you open the doc, you will see the endpoints and the exact request that the tool accepts. Not every tool accepts every request. So your ability as well is to kind of find out what that is and so on. Again, you can use AI for this. Like you can use AI. Cloud is very good. If you just feed the the link of the API documentation, it can just look it for you. But if you understand how it goes and how it works much better. So how does an API

[21:14] documentation actually look like? Right? I'm going to use three examples. Instantly, which is a cold demo platform, Apollo, which is a lead enrichment platform, and ClickUp, which is some sort of like project management CRM kind of platform. Right? So the first one here is instantly. I'm actually going to go to instantly's documentation and show you right there. So I just searched up instantly API

[21:31] documentation and I go here. All right. So here is the API documentation. On the left hand side is where you can see all the different actions that you can do with the API of instantly. What can you automate? So there are three different components of an API documentation. There's the actions, there's a description, and there's a request. So the actions is essentially all the actions that you can do with the API of instantly. What are some things you can automate and cannot automate um instead of instantly? Well, I mean it doesn't really tell you what you cannot automate, but if you can't see it here, then you cannot automate. You can also

[22:03] use instantly feature, you know, in case you want to ask it any questions. So, for example, let's say we want to ask it, hey, I want to send a test email

[22:16] using the AI assistant. Now, it will look through all the API references. It will look through everything and then it will give me it should give me this right here a description of how I should go around you know actually automating the process of sending a test email. So using a post API v2 emails test which is the one here with the fields that we have here and so on. We have the request here all these and then we have the description in the middle which is the description of what it is that you're doing. So send a preview test email without creating an email entity in

[22:44] unibox rate limit 10 requests per minute per workspace. Right? And so this tells us what it is that we're automating. And here it gives us a ton of information as to like what the request needs to include. So it needs to have authorizations. It needs to have like different variables that we send it to. And that's the description. That's the that's the part in the middle. On the right hand side, we have the request. So this is where we get in the depth of like a curl. So a curl is basically a piece of code, which by the way, you don't really have to understand fully,

[23:07] but it's essentially how the AI is able to understand what thing we want to automate in the software itself. So when we go into like actually building this inside of N10 and retail and and clock code, we'll be able to just copy this and paste it into the software itself and it will set everything up for us, right? So if you're wondering, hey, I'm

[23:23] using instantly, how do I delete emails? You go here, go down, delete emails. And then you'll find that this is the endpoint, which is kind of the the URL that we call saying, hey server, I want to do this. And then you see all the parameters and then you basically see how you set things up, right? And I want to go back to here, send a test email because we can see in the curl request that we have post because sending a test email sends you're sending something over. But if you go to get email, you can see that this is get. And you can also see this right here. Post, post,

[23:52] post, get, get, delete, patch, get, post. So that's the difference between post, get, and patch, right? You're basically doing different actions. So sending a test email would mean that we have to send a post request to a server saying, "Hey, I want to send a test email." But if you want to use a get email, then we send a get request. If you want to patch email, so update an email, then you will use a patch request. And that is how it applies to each software specifically. So with that said, that is the API documentation of instantly. On the left hand side, we have the actions. In the middle, we have

[24:20] the description. And on the right hand side, we have the request. If I go to the Apollo API documentation, you can see that it's pretty much the same. On the left hand side, we have, you know, enriching people, organization, enrichment, create an account, update an account, search for accounts, which are actions. In the middle, it's the description of the action. And on the right hand side, it's the request which is the curl. So post request with this URL. You need to have this header, this header, this header. And these are the things that we can do. And finally, if you look at ClickUp API documentation, again, it's pretty much the same. So we

[24:48] have the actions on the left hand side. So things that we can do. You can press in each one. In the middle, you get the description as to what the endpoint is, but also what is it that we're doing and also what things do we need to have in place to then send the request right here. Now I did mention earlier how AI can also help you to be able to use the API documentations and see whether you can automate something or not automate something. So let me show you. So right here I'm in chatbt and I can say hey I'm I'm using instantly in my automation

[25:16] inside of NA10 [music] and I need to create an API request. Now the action that I want to do is that I want to send a test email. Can you look at the API documentation of instantly and tell me whether it's possible? If it is tell me how to set things up step by step. So I'm going to go and press go. And now what it should do is it should look at the internet look for the API documentation of instantly. And then it should give us if it's good enough a response with how to set things up. All right, there you go. So it gave us the

[25:42] actual uh response itself and now we can fact check. So we can see that it says method post. Method post is fine. So that's post. All good. The URL is https API instantly API v2 emails test. API v2 emails test. That's great. The reason why this first part isn't there is because we're looking at the official instantly documentation. So the the first URL is not there. Authorization this and this match. And by the way, this is a bit confusing. I'll explain exactly what each one means later. And then we have the JSON. So account to address subject body HTML. Account to address subject body HTML. Right? And

[26:15] then you can also ask it questions like, hey, I don't know what any of this means. Can you explain to me what what it is? Right? So now you got to see how we can also use AI to then go inside the API documentation and see whether it's possible for us to do a specific action and how to set it up within our automations. So those are three examples of API documentations. So you get an idea of how it looks like and you can see that it's it's a very common pattern around each one. Three different things, right? Actions, description, [music] and request. And so if you understand that

[26:44] pattern, it's going to be much easier. And the way to get better at this is to just practice. And when I say practice, I mean just watch this video when I build things and you will see that I have to go to the API documentation of a software to get the request and set things up. And by the way, it's super super easy. Like much easier than you think. All right, so with that said, here's exactly what to remember from the foundations of AI agents. An AI agent can think, decide, and act. A chatbot just talks to you. An AI agent is consisted of five parts. We have the

[27:10] brain, which is the LLM, how it thinks. Instructions, so your prompt, your instructions as to what you want it to do step by step. the memory which is how it remembers the conversations, the knowledge which is documents and data that you give it for it to always retrieve to answer questions and then we have the tools which are the softwares

[27:26] that it uses to take the action. Workflows are fixed and predictable. AI agents adapt. So workflows are a series of steps that happens every single time. Hence why we call them deterministic and AI agents are non-deterministic because one input can lead in different kinds of outputs because the AI agent can think. And then APIs are how agents act. And we have post, we have get, put, patch, and delete. All right. So, congrats on finishing module one, which is the foundations of AI agents. You now understand exactly what an AI agent actually is and how it works, the different components of an AI agent, and then finally, how AI agents are able to

[28:01] speak to software through what we call APIs. And I can tell you that you are now ahead of most people inside the AI agent space because you understand the foundations of how everything works. Now, there's one quick thing before we keep going, which is the fact that everything I'm going to talk about in this video and the things that we're going to build is exactly what my

[28:18] clients and me do every single day. [music] So, if you want my personal one-to-one help to be able to take these skills that we're going to talk about and actually turn them into a real business onetoone with me, then I left a first down below in the description which walks through a full video on how you can do that step by step. There's also a free school community where a lot of this is already laid out in case you would rather start there. With that being said, let's go to the module two,

[28:41] which is the building. So, in this chapter right here, we're going to be building five different use cases using four different agentic platforms. And I'm going to cover pretty much the main sort of use cases that you can actually start selling to businesses. So, the first use case is going to be a website lead generation agent. So, this right here is going to be primarily built with voice flow. And I'm going to show you how you can turn traffic from websites into actual leads that the business can close. The second use case is going to be an inbound receptionist agent built primarily on resell AI showing you how you can uh

[29:14] pair up voice with booking appointment systems to create an agent that can answer the phone 24/7. The third use case is going to be a speedtole agent. And in this use case, we're going to build an agent that can follow up with ad leads within 5 minutes of them opting in to increase conversions uh for the business and make them more [music] money. The fourth use case is going to be an AI executive assistant agent showing you how you can build an [music] agent that is able to automate different parts of the admin plus operations for a specific business. And the fifth use case is going to be a sales lead

[29:45] research co-pilot. And we're going to be primarily building this with clock code showing you more advanced custom agent building. And this is for sales teams to be able to have as much information about a prospect before they jump on the call and sell the prospect. Now, as you're watching this video, it's also important to understand that you pretty much learn the best when you're also doing it with me. So, the first time you can watch me do it, but the second time when you rewatch it, I recommend that you do everything step by step alongside me because you learn by doing. You make your own mistakes and you learn along

[30:15] the way. With that being said, let's go to the first build, which is the website lead generation agent. All right. So, this agent right here will be living on the website and it will answer certain questions and capture leads which is extremely important for businesses that have a ton of visitors in the website but have no way of converting them. The first thing is the visitor. So, the visitor lands on your website. Let's say that's you. You go on a website, you're interested, you will open up a chat widget which is on the bottom right. Then you'll start asking questions to the actual agent

[30:43] which will be hosted on voice flow. That's the platform that we use. And then voice flow will have access or the agent in this case will have access to a knowledge base. It will have access to tools which we will build together uh to be able to answer any questions and take action. And finally once it qualifies the actual person itself and sees that hey they're a good fit to work with us then it will capture the details uh the name the email and the need itself so that our team can then contact them later down the line right and so how is this better than the current process

[31:10] that a business has well a lot of businesses like ski resorts hotels which has a lot of visitors that go on the website these people usually either have a very very old chatbot or don't have one at all and so they have no way of qualifying the person they have no way of answer any questions and they have no way of capturing the leads which is a lot of missed opportunities for each kind of business and each opportunity is an amount that they could have made of the business right all right so the first step is signing up to voice flow so here we have voiceover.com you want

[31:37] to go to the signup page you want to uh fill out the details with your email or Google I sign up with Google so once you have sign up to voice flow you actually do get some credits that they give you I personally upgraded to the $60 a month plan because I'm able to have different flexibilities and there are some features that we only can use if you have the paid fun. So, I recommend you get that if you're actually going to be using this for your business or to

[31:59] implement in someone else's business. Bear in mind that when we work with clients like they're the ones who will get the voiceful access and they will add you under the members, right? So that you can access their account from your email. Very important. So, we have home where you can use AI to help you build anything. Uh then you have projects. So, this right here will be where you can see all the different agents that you're building. In this case, we have none. So, we're going to build one uh in just a bit. We have usage. So this is uh looking at how many credits you're currently using. So I

[32:28] used 1% plans and billing. This is your plans and billing. Um you can upgrade it, downgrade. It kind of depends on how many businesses you're implementing this for or the complexity of the system or how many users are using this chatbot uh or this chat agent, not chatbot. And then we have members. Again, this is where the business itself will give us access because ideally we never want to use their account specifically. If there's a way of us getting access through our email, then that's awesome, which is what they do. Um and then phone numbers because recently voice flow implemented the ability to create voice agents as well. Right for voice agents I

[33:00] don't really use voice flow. I like to keep it on chat for now. Uh we use retail AI to build voice agents but that's something you can do here as well. And then we have a bunch of resources here like the documentation the API reference tutorials support and sales all that stuff which you can do as well but these are the main ones. So the first thing you want to do is you want to go to projects and you want to press new project. So it will ask you for the name. In this case, we can do ski resort agent. And it will ask you whether you

[33:27] want to do a web chat or a phone call. So this is the difference between a chat agent and a voice agent. In this case, I want to stick to web chat, which is the agent itself, uh like a chat agent that we're going to use and implement on the web. And in here, we can see that it's asking us for the framework. So we have the ability to do a gentic or conversational flow. Now bear in mind, I started working with voice flow back in

[33:47] mid 2024. didn't have Agentic, hence why it's relatively new, right? Um they only had conversational flow. So the difference between one and the other and the reason why we're going to use Agentic is because conversational flow it's sort of like workflow automation platforms like you have to manually build each step like step by step, right? Whilst Agentic, you give it a prompt with instructions and it will kind of make its own flow based on your instructions which is better. So we put Agentic. Um and it's also more kind of free spirit per se. Objective. Um, yeah, leave this as is. Just put start from scratch and

[34:22] we can start with the agent itself. All right, so we are on this page right here. Now, if you're new to voice flow, this might seem a little bit intimidating, but it's actually very, very easy once you break it down with a few things cuz it's not much stuff that you have here. The first thing is the AI that it's going to use. So, if you remember before when I mentioned the foundations of AI agents, I mentioned that the first thing is the LLM. It's the brain. [music] So, the brain is this right here. Now, as you can see when I zoom in, um, there you go. On the right hand side, it tells

[34:51] me, hey, look at the speed, look at the cost, and look at the intelligence. And so, ideally, we want something that's like pretty green on all three. I found that GBT 5.5 is one of the best when it comes to this cuz the cost is fine, the intelligence is pretty good, and the speed is pretty good. So, you want something in between. Now, bear in mind that as of the time of this recording, 5.5 is your best bet. But with everything in the AI space, the more the more time we go on, the more models come out. And you'll have different options here that you can see. So now that we

[35:18] chose our model, which are going to be 0.034 per message, like the pricing model of of kind of how it works and the speed. So now we have to create the global prompt and the instructions. So in this case, the global prompt is the one that's applied. So every turn to shape how the agent responds independent of the rooting or tools. So this is the kind of the main thing. It's like, hey, you were this agent. Your goal is to do X, Y, and Z. This is giving it the identity. is sort of like the system prompt and then we have to do the instructions which defines how the agent

[35:45] routts requests how it decides which tools to use how it basically handles what it needs to do. So when this happens, do this. When that happens, do this. Right? And so it's a mixture of different things. And once we have these two, we can go to creating the playbooks, which are sort of like the sub agents. I'll explain what that means, don't worry. So I'm going to go here to my AI agent masterclass resources, which by the way contains all the prompts for all the builds we're going to do today. And you can find the resources in the third link down below in my free school community. You can go

[36:11] to the resource section. You can go to the templates vault and you'll find everything there. So the first thing of the agent is the LLM, which is a brain. Now we get to the instructions which is your stepby-step uh kind of plan, right? So here's a few things about the prompt that you can see off the bat. So the first things first is that we give it the identity. So we're saying you are XYZ person because the way that AI works is that if you tell it who it is, then it will basically like trick itself into

[36:35] thinking that it actually is that thing. It will embody the identity and it will give you a much better answer if you're asking it questions specifically towards that kind of identity, you know. So if you say you are a virtual assistant and you ask it questions around you know virtual assistant then it will give you a much better answer right the second thing that we give it uh in this case is going to be the tone of voice so warm approachable uh enthusiastic natural conversation language that's fine we give a response rules so keep responses concise one topic at a time use natural acknowledgements great questions

[37:04] absolutely so this is basically the different rules throughout the conversation and the last thing here is guardrails now guardrails is essentially different rules as well but there are things that we give AI for it not to go off track. So we say never discuss competitor ski schools, never provide medical or safety advice, never confirm bookings, never share instructive personal data, do not discuss internal business operations. If a guest is frustrated or the issue is complex, escalate to a human team member. These are different rules that we want the AI to implement. Now, the best way that I

[37:31] can think about guardrails is bowling. So if you guys have been bowling before, you take the ball and you try to hit the pins and get them off. If you're anything like me, I'm complete ass at bowling. Like probably one of the worst you've seen so far. So when I go to the bowling alley, I'm normal, so I don't I don't actually just strike it with nothing in between. I use the

[37:47] guardrails. So I put it up, right? Because I can't hit the balls for my life. And so those guardrails allow me or allow the ball not to go to the side and it hits the pins. Yes or yes, which is how I win all my matches. But that's the way that I think about guardrails is different things that you put on the sides of AI. So it doesn't actually go beyond the thing, right? It just it stays within the track itself. Hope that was helpful. I hope that analogy helped to understand guardrails. So that's the global prompt. So what I'm going to do is I'm going to copy this and then I'm

[38:15] going to paste it here. I'm going to just go full screen and you can see that we have the prompt here. So again, we're not telling it exactly what to do. We're telling it who it is, what the main kind of concept or the thing about the agent is. Now we go to the instructions. So this is telling it, hey, when this happens to this, when that happens to this, right? I made the prompt for this as well. So we can go here. I'm going to go to instructions and we see here. So, conversation rooting. Root the conversation based on what the guest needs. And we have four

[38:41] different things that the agent is able to do. So, we have booking and pricing. Use the booking playbook when the guest wants to book lessons. Ask about pricing, lesson times, group sizes, or availability. Resort information. Use the resort information playbook. When the guest asks about meeting points, bus routes, ski pass information, and so on. Instruction, use the instruction playbook. When a guest asks about XYZ, general FAQ, use the main support playbook uh for general questions about Tori ski resort, what to expect, what to bring, cancellation policy, weather, or topics not covered above. These are instructions, right? As you can see, this is not a massive prompt. It's just

[39:16] one of those things where like it's saying, hey, categorize the response of the user to say, hey, when the user asks me about this, then send it to this this way. When he asks about resort information, then send it to this playbook and that playbook. Now, your question might be, what is a playbook? So before I explain what a playbook is, I'm going to copy this and then I'm going to go back to voice flow and I'm going to paste it here. And again we

[39:39] have starting message we have the tools. Now one thing I forgot to mention as well for the prompts. If I go to global prompt you can also see that we have these hashtags. The way that these hashtag works is that if I go here these are headings. So this is heading two. If I go here and put one hashtag it be heading one. If I put three heading three heading two and so on. Right? you can see the gist of things. The reason we do this is because AI just like a human needs to categorize different information. It needs to know that this is a title of this and this is a title

[40:07] of this. That way it knows that this information just corresponds to the response rules cuz if everything was the same font then it wouldn't really know what's more important like what is what essentially. Now to the point of playbooks, right? Playbooks are what we have here. So these are essentially sub agents. Now sub agents are are an extension of the main agent. So the main agent like the the boss of the company has different employees, right? And says, "Hey, if someone asks me about XYZ, I'm going to send them to Sarah. Or if someone asks me about resort information, I'm going to send them to Jimmy, right? And Jimmy will answer any

[40:37] questions about the resort information." And so we want to have different playbooks for different things, right? So now we have the global prompt, we have the instructions. Now we have to create the different paths of the actual uh where the agent goes, right? In case we ask any questions. So first things first, we go here, we press plus for

[40:53] playbook. I already made one just now. So we can just press create playbook. The first one here we can name it main support playbook and then the trigger we can just say use this for general questions about touris resort what to expect what to bring cancellation policy weather topics and and so on right then what we would do is I would paste the whole prompt that I have here right here. So here's a prompt we say um main support playbook your job the knowledge base. So here's all the questions that it could answer. Again, this would be like the document that it uses to to be able to answer any

[41:26] questions. Lesson types, lesson times, meeting points, what to bring, ski pass information, bus transport, cancellation policy, conversational flow, greet the user, answer route, suggesting steps, and so on. So, these are all the different things that it has to do inside the conversation. Button strategy, rules, response guidelines, example, escalation, and so on. When we're talking about general questions, it will go down this exact like route of conversation. Okay? We're not going to use any tools for this because it's just about answering questions. And this will be the knowledge part of the actual agent. So we have the LLM, we have the instructions. This is the knowledge part. So if you ask any questions about

[41:58] this, it will pull the information here. Lesson types, lesson times, meeting points, what to bring, and so on. That's the first thing. Now, I do also want to um press here and give it a I think it already gave it to me. Yeah, it's fine. There you go. Uh this is a trigger. So saying, hey, trigger this when this happens. The more specific you are about what what the trigger is, the better the result will be. like the more accurate it will be sending it that way. That's the first playbook. So that will be the main playbook general FAQ. The second here will be instructor information. So

[42:29] instruction playbook. So create playbook instruction playbook and then the trigger will be use this when the guest so use this when the guest asks specific questions about instructors. [music] instructor qualifications or request a particular instructor and then I'm going to paste this old prompt here. So this whole prompt will give the instructor information. Your job is to provide general information about our instruction team. Help guests feel confident in the learning with experienced professionals. [music] Instructor request should always be treated as a preference rather than guarantees. Right? Our instructors and then we go into Ethan is this. Isabelle is this Luca is this? Elena Ryan and so

[43:12] on. And then some more information as well. Uh and guard rails. We always want to add guardrails into the prompt to make sure that you know it doesn't actually deviate. Uh so that's the second one. Then we have to make the third one which in this case is resort information playbook. So resort information and we're going to write use this when use this when the guests ask about meeting points, bus routes, ski pass information, resort facilities or getting around. And then I'm going to be pasting this whole prompt here. So the

[43:43] prompt says pretty much the same thing. Again, all the different formats and stuff, they're all pretty much the same. Like they don't really change. We have medium points, we have private lessons, ski pass information, transport, and so on. Lockers, common questions, and so on, and everything else. What you might ask, well, if these are all like answering questions, why don't we just have like one prompt with all the questions and everything there? It's because we usually like to split things up in terms of the context of the question. So if this is just a general FAQ, then we much rather have a playbook just for that. If it's about resort

[44:14] information, [music] we have a playbook just for that, right? So we rather have one like different employees for different things, for different questions, types of questions rather than one employee with all the questions because it can be inaccurate. It can use different answers to the wrong questions, you know, and splitting things up gives the AI more space to breathe and it will increase the accuracy of the agent itself. Now we made the third one. Let's do the last one which is booking playbook. Um I go

[44:40] here plus create playbooking playbook. Now I'm going to say use this when the guest wants to book a lessons ask about pricing lesson times group sizes or availability. And we go to booking and we can paste this here. So you handle lessons bookings and pricing increase for tort ski resort. Man, I wish I had a business with that name. Your job is to help guests explore lesson options and so on. group lessons is this, private lessons is this, mountain kids program, multi-day packages, session times, and we have a booking flow. The booking flow is essentially the the conversation that it goes through to ask qualifying questions. So, how many guests are

[45:17] participating? How many adults, which dates, uh what is your ski or snowboarding experience level? Because then it uses all this information to then send to our team and we can then follow up with them because our agent already had a conversation with them and it shortened the time of us reaching out to them because we already know what

[45:32] they want, right? It's much easier. Okay. So, um, gather information, make a recommendation, gather booking information, send to reservations. So, pass the details into the booking workflow. Inform the guest. Our reservation team will send you booking confirmation within 2 days. Payment instructions will be included in a confirmation email. There we go. Okay. So, now we finished writing the

[45:51] different playbooks that we have here. So, in this point, the agent is now able to actually use this playbooks to answer any questions specific to what the user wants. All right now, I'm going to turn down cards, carousel, web search, and uh [music] well, and is fine because we don't really need them. I feel like we don't even need knowledge base because knowledge base is included inside the

[46:09] playbooks. So now I can just press run. I can test this agent right here. You're going to see a whole line of code and this is the logs. So this is when I when I ask a question like I'm going to ask any question specifically about the resort or whatever and we can see how the agent actually acts. So I have a

[46:25] question around this resort information. So, it'll use the resort info playbook because we asked it something about the resort information. Meeting points. Yeah. Where are the meeting points? Adult, kids meeting point, private lessons. We can do kids. The kids meeting point is outside the ES building about 50 m the Shaane entrance. Okay, there you go. Um, I want to book a lesson. And you can see here how it now went to the booking playbook. So, this is cool because especially when it comes to debugging, like making sure that things are working right, we can see what went wrong based on what it used, right? Uh, I can help you find the right

[47:03] lessons. How many guests will be joining? Um, five guests, two children, three adults, three audits. Okay, my bad. Um, which dates are you interested in? Uh, we can do let me see if if this works because I know AI is pretty bad with dates. So, let me just say uh we can do tomorrow, which is 25th of June.

[47:25] So, what date is tomorrow? Okay, perfect. It actually is pretty good. Um I believe because we added a date to the actual prompt. Anyways, uh what everyone's experience level, we are all beginners and can do group five beginners with a group. Okay, perfect. for morning session. Morning morning pass information. Please send full names of all private participants. Okay, so it asked me for the names, ages, email address and phone number and specific request. I just put Jack Jane Kelly Luca James and surname is Daniels. One is 10

[48:01] years old and the other is 8. John@gmail.com 1 2 3 4 5 whatever. And then no special request. I'm going to press this. And now the agent itself will retain the information. Right. So now it says thanks for the details contact and so on. Our reservation team will send you booking and confirmation within two hours. Payment instructions will be included in the confirmation email. First things first, this is great for testing just because of the fact that like we can see all the logs. So

[48:25] the whole conversation is here, right? The conversation started, the main agent started. How can I help you? We said, [music] I have a question around resort information. We can see that the main agent transferred to the playbook resort um info playbook. Then we said, what would I like to know? Asked where are the meeting points? it uh answered the question and then I said I want to book a lesson. There we go. I want to book a lessons and then the resort info playbook called transfer to playbooking playbook. Meaning that now we go from the resort info playbook to the play uh booking playbooking playbook playbook,

[49:00] right? I should have changed I should have changed the name but there you go. So we go from resort to booking, right? And so we can see that different playbooks are used based on the context of the conversation. Uh which is perfect. And by the way, you can also see the the pricing here. All right,

[49:11] cool. So press X and we go back to here. So the testing is good. Now there's one thing that I um didn't cover here which is the knowledge base. Right now I said what it is but I didn't actually add anything there. [music] And the reason that is is because we already added the Q&A on the playbooks. So here the main support I already have all the information. The knowledge base is already here right? So you can do it either way. The only time where I go here to knowledge base and ask and start adding stuff is if it's something that's super super long or if it's a web page

[49:42] right now. How does this work? So you can actually put a link to a web page and it will scrape the web page meaning that it will extract the information from your website to answer any questions. So I can go here to webpage and we can actually add multiple web pages. So, uh, front desk as a IQ.com. Front desk IQ. So, I made a website here

[50:04] called frontiq.com. And the way that this will work is that you will basically just copy this. Go back to voice flow. You can copy this. And then the refresh rate is how often do you want the agent to rescript the whole website. The reason this happens is because a lot of websites change. You

[50:22] might add a about me section tomorrow. You might add a I don't know another service, right? And so you want the agent to like continuously be updated with the information that you have on the website. Then you have LLM chunking strategy. Um you can just do FAQ optimization because we want to optimize for question and answers. That's fine. And then folder is it's okay if you leave it as all data sources. And then you also have metadata which I don't personally use. Um so you can leave it at that. And then you can import one web pages. Now I'm not going to use this because this website is completely off

[50:50] off the radar of whatever we're doing here because it's a ski resort. This is an MR receptionist agent. So it's different. [music] So with that said, you can also add web pages, you can add site mapaps, you can add docs, tables, plain text, zenesk, and Shopify as well in case you want to connect this

[51:04] knowledge base to your Shopify store. And this is a very very good use case because a lot of e-commerce stores have Shopify stores and they keep adding more and more products and so the agent is always updated. So at this point, we have finished the global prompt, the instructions and all the playbooks. But the next step is actually making sure that when we want to book an appointment, we don't actually book it here. It says it will keep the details and then let our team know. And so we need a way for us to transfer information from voice flow to somewhere else. So in this case, we can make a

[51:33] Google Sheets here. I just put sheets on you. And we can name this ski resort lead information booking. I'm going to leave the headers blank for a second because I don't know what I'm going to put there. But if I go back to voice flow and I go to the booking playbook for tools, I can connect it to a Google Sheets through API. Right? So it has an API. I can press on Google Sheets. All right. So once you connect to your Google Sheets, you'll have this right here. It will say connected. Now I actually have to switch my browser cuz I was using Firefox, but it didn't work

[52:07] for some reason. So Google Chrome works fine. Once you connected, you want to press add to sheet right here. And then you will be on this page. Uh you can go to spreadsheet URL. Select a spreadsheet. Make sure it's your account that's actually connected. And then you can go to ski resort. Ski resort leads information. There you go. Select

[52:31] uh value provided. We can go to range. We can leave this as is. You can do sheet one and then values. So here is where it gets just [music] a bit tricky, right? Because the values I'm going to paste this here. So what what is this that I put here? Looks very complex. It's actually pretty easy. All my Google sheet that I made here, I actually added different names of columns. So I added lesson type, activity, level, time, date, number of guests, participants, email, and special requests because this is the place where the agent is going to add the information to when it receives all the requirements from the user. All

[53:02] this is saying is output exactly one array containing one inner array of 14 values in this format. V1, [music] V2, V3 until V14. Do not create a a separate array per field. Do not include any fields, label or col names, values only. And then it will tell it to give me everything in a specific format in a way where it can fit towards this. Right? So I recommend that you just like copy this. You can take a screenshot, put put it throught say transcribe and you get the text. And make sure though that these are the same names. Right? And you can change the names as well. We've now

[53:33] implemented this as well to make sure that when we do talk to the agent, it actually sends all the information correctly into the Google sheet. And that's it. So we should have this, this, and this all good to go. And now we can test. So, we can go here, run, and we can go to I want to book an appointment. And now it should ask me a couple questions about me specifically. By the way, we should see the information coming here after we give all the information. So, it will be five guests,

[53:59] uh, two kids or actually all adults. Dates we can do tomorrow. Beginner snowboarding. Okay. What times would you prefer tomorrow? Morning is fine. Perfect. Tomorrow morning to pass this information. Full names. Jack, Daniela, Luca, James, and Christian.

[54:23] Email address. So, ages uh 20, 22, 24, 52, 50, and 64. My email address is going to be john@gmail.com. The most common email on the internet. I swear whoever has john.gmail.com is actually just getting spammed all day by people trying to test agents. And then we can say no special requirements. There we go. Then asking, should I use

[54:50] Tory as a surname? Yes, use Tory. As you can see here, let's scroll down because here the agent booking playbook called the Google Sheets append spreadsheet tool and it succeeded. So if I go here and refresh, we can see here that we have private lesson request tomorrow morning 9:00 a.m. We should also add the date of today in this case. Um time date. Okay, so the the colums are a bit messed up which is fine. We can just update these. Number of guests, participants, names, special requirements or in this case ages and then email. um no special requirements and 210 bucks an hour which is the

[55:31] price. So you can obviously go ahead and update this but you got to see how the agent is now able to access Google Sheets through what we call an API and tools, right? All right. So I just updated the actual columns so they're all accurate. So then next time it actually runs it runs successfully. All right. So now that we finished the agent, so we have the global prompt, we have the instructions, we have the playbooks, we have the API, we have the tools, we have all that stuff. The agent is finished. What happened to the agent now? How do we publish it? How does it

[55:55] work? So ahead of this video, I actually made this website right here. This website is just a very regular website that I made with claw code where I wanted to show you how you can now take the chatbot and then deploy it on the actual thing itself. So when we go here, we can go to publish, press publish here, and then we should go to widget, which is a deployment. And this is the important part. So I'm going to copy this code right here, which is going to enable claw code. I'm going to say hey claw code put this code inside the website because the claw code is the is

[56:23] the software that made my website and does all the code and it's going to take this it's going to update GitHub which is a place where we store all the different folders of the website which will then update my website hosting platform which is Versel. So super super easy but I'm just going to copy this and I'm going to go here to cloud code where we have the travel agency website and then uh it says paste this code snippet before closing the body. So I'm going to copy this. Copy this first of all and paste it here. But then before this I'm

[56:53] going to copy this and paste it here. And then I'm going to say I built a chatbot or an uh chat agent using voice flow. Please follow this instructions to deploy it here and push it to GitHub. I'm going to press bypass permissions. And don't worry if you never used cloud code before like we'll talk about this as well. Um I want to show you the whole process of of updating it though. says, "I'll add the voice flow chat widget script right

[57:18] before the closing body tag." Perfect. All right. So, it finished updating it to GitHub, which is going to be the one right here. All right. And on the website right here, we can see that we have deployed the agent. We can go to talk to AI. Uh it will ask us, "Hey, welcome to the tortoise ski resort. How can I help you with your ski school

[57:33] plans today? When are you guys open?" So, not not the best spelling, but um it gives me the answer, right? So we basically deployed it for this website specifically and now we can go ahead and obviously change the logo, change the name and so on. But that is a full process from building to actually deploying this on the website itself. To change the appearance, we can just go to here obviously change this like the different dimensions. But then for appearance and style, we can change the logo. We can change different uh the header like the way that you know the actual name of the agent and other

[58:06] things like that as well. So there's a lot more things that you can do with the AI agent itself. But you got to see how we were now able to build the agent and then deploy the agent on our website as well. And when it comes to delivering this for clients, the actual deliverable that you'll give for them is the the code, right? So you give them the code, not them specifically, of course. They wouldn't really know what to do with it, but typically the clients that we work with have someone that has built them a website. And so if you tell them, hey, uh, paste this code snippet for before

[58:33] the closing body and you give them the code, they will do that and it will automatically update on the website. And so that's the easiest way that you can deliver this for clients. Or if the client doesn't have someone who's responsible for the website that you can just hop on a call with the client, you can do it for them, which we've done before as well to make it as easy as possible. And we're going to go through the whole delivery process, packaging, pricing, offers, how to sell these to businesses as well at the end of the video. So stay tuned for that because it's going to be good. All right, so

[58:57] that marks the end for build number one. Now, bear in mind, I'm going to give you guys the blueprint of the actual agent. So, you can just go here and you can press import and you can import the file. You can find everything in a second link down below in my free school community. You can go to the classroom section and you can go to the templates vault. And now that we're done with build number one, we can go to build number two, which is an inbound

[59:16] receptionist. All right. So, the way that the inbound AI receptionist works is that we're going to build a voice agent. So instead of doing a chat agent, now we're going to build an agent that we can call over the phone. And this agent is going to be able to answer calls, book appointments, and answer questions. So it all starts from the user calling the actual business. Then we're going to use retail AI as the platform to build uh the agent. Retail AI will have a knowledge base, which is a document that the agent uses to be able to answer questions. And then we have cal.com, which is a

[59:46] platform that allows us to book appointments. And obviously uh the caller is is finished, right? Handled. So you'll be an agent, book appointments and answer questions. Very, very simple. Um, it is all going to be built on retail using cal.com as well as a platform to book appointments. So with that said, let's dive in. All right, so the first thing you want to do is you want to go to retailai.com and you want to go to try for free. Once you press try for free, you'll get on this page. You can press sign up and then you can use your email. I'm going to press continue with Google. You can

[1:00:13] add your code and then finally you get into your account. All right, so once you get into your account, you will get on a page that looks like this. Uh this is right here is the retail dashboard. So we're going to go through everything step by step. If you've not seen retail before, it's actually very very simple to use. Hence why I like using it. It's super super comfortable and very beginner friendly, especially for people

[1:00:30] who don't know how to code like myself. So great. Now the first thing we have on the left is the the different options, right? We have build, we have deploy, we have data, we have monitor, and we have system. All right. So the first thing we can see here is we have agents. This right here is going to be the place where we build the agent itself. The

[1:00:45] prompt, the settings, all that stuff. Then we have the knowledge base which is where we're going to be adding uh as you can see here like a PDF or some sort of like document that the agent is able to use to answer questions. Then we have phone numbers which is where we can buy the phone number. Now bear in mind that you need to verify yourself which we will do in just a second. But if you're in America I recommend buying in retail AI because it gives you the ability only to buy US numbers. But if you are in

[1:01:11] Europe then this wouldn't work for you. you have to go to Twilio or Telnix or you can use other providers as well uh to buy the number. But again, this only allows you to to buy the number in the US. Then we have the option to batch call. So call um multiple people at once. We have call history. So you can see all the different calls you've done before. You can see chat history, uh contacts that you have in case you want to add contacts. You can sync your CRM into retail AI. You can monitor your

[1:01:35] calls. You can monitor the live calls. You can do a quality assurance, alerting, integrations, billing, and settings. Right? There's a ton more stuff that we have here, but the main ones that we're going to use is going to be agent, knowledge base, phone numbers, and we're also going to see the calls that we had using our call history. So, with that said, let's go here to agent and press create an agent voice agent and make sure that you don't press conversational flow, right? You just press single prompt agent. And the reason why we want a single prompt over a conversational flow is because with a single [music] prompt, you're able to

[1:02:07] have what we call a free form conversation instead of a deterministic conversation. So what's the difference? A free flow conversation is sort of like a conversation that can go any different way depending on what the user wants. A conversational flow, a deterministic conversation is a conversation that you can determine. So step one is this, step two is this, step three is this. And it

[1:02:27] doesn't really go beyond the guardrails. It just stays there, right? And that's not what we want to do. We want to have a conversation with the user. So you can put single prompt. You can put create an agent. Once you create the agent, you will get on a page that looks like this. All right. So this right here is the retail dashboard is where we're going to build the actual agent itself. Choosing the, you know, the welcome message, the AI, the cost, the latency, the voice, the language, the function, knowledge base, all that stuff. We can publish it, we can test it. Everything is here. So

[1:02:54] super super simple. We have on the left hand side where we're going to be adding the prompt and we're going to be changing the settings in terms of AI and voice and language, right? Of course, language is going to be English, but then we have the AI and we have the voice. Now, the way that the agent charges you is by minute. So, we have the cost and the cost depends on what AI you use and basically what voice you use. So, in this case, the voice is connected to a software that we call 11 Labs. So 11 Labs is is a software that has different voices and retail AI has a

[1:03:24] partnership with 11 Labs and we basically use the voice in 11 Labs for our agent. So if you go here for example and I press play. Hi Simo, an AI voice agent for Retail AI. My calm friendly voice is here to guide you through anything you need. There you go. So we have different voices. Even this guy is a guy customers

[1:03:44] money. There you go. So you have different voices. Um, I'm going to go with Simo for now just to keep it very very simple. But you can go around and and kind of choose which language you want. So those are the voices. I recommend that if again if you're in America, this will be great for you as in you can use all the voices that are here because primarily English is the main language that people use. But if you were in Europe or any different other place uh that's not English language, then I recommend using fish audio. Okay, it's a new software. I would recommend using

[1:04:09] that um to to choose the voices. This is the cost again based on the AI model and the voice itself. And if I change the model to let's say GPT 4.1, you can see that the price changes. So the model actually does impact the price. I'm going to choose 5.1 for now. It is on the cheaper side, but also very very good in terms of quality. Then here we have latency, which is the time between when the agent speaks and you speak and kind of the conversation and how natural it sounds. Then we have tokens, which is the amount of tokens that the AI uses or

[1:04:36] that you know AI is able to use to in a conversation. And you can see here that prompts exceeding 3,500 tokens incur a supercharge. So, you want to keep your prompt uh 3.5K tokens below to make sure that you know it has all the information. It doesn't it doesn't overload. We have the AI, we have the voice, we have the language. You can choose whatever language you want. I recommend that if you do Italian or, you know, French or whatever, your prompt should be French as well, cuz then it's going to be easier for the AI to to understand. Then down here, we can see the welcome message. So, in this case,

[1:05:04] we're going to be adding a welcome message. So, we're going to say the AI speaks first. Here I'm going to put custom message and I'm going to say hello this is Grace a digital assistant for Tory Law Estate. How can I help? So in theory now whenever we press run test the agent will say this at the start. So whenever you call the agent this is the first thing that the agent says. Okay let's let's try this out. Now, there is going to be an issue though because when I press run test, it will say the sound out loud and it will go inside my mic

[1:05:35] and it will make it seem like I'm talking back to the agent. But we'll just test this. This is Grace, a digital assistant for toy. Okay, there you go. So, you saw that it said the thing, then it was going to go back and have a conversation, but that's not what we want to do. So, that's the first thing. Uh, now the second thing is going to be adding the prompt. So, I already have a prompt here. I'm going to paste it here and we'll go through everything step by step. So we have the role, objective, personality, context, instructions, we have available tools, we have end call, we have qualifying

[1:06:04] questions, we have qualifying criteria, we have stages, um and then we have example interactions. Okay, so super super easy. The prompt does follow a very very strict like framework that we use for every single prompt for voice agents just to keep them as yeah simple as possible and as structured as possible as well. So the first things first is a role of an objective. So we

[1:06:22] say you are grace a digital assistant. Your primary objective is to qualify incoming callers and book them for 30 minutes consultation if they are qualified. Personality says you are warm and reassuring without being exaggerated. Emphasize your warmth when the user expresses emotional distress but efficient when they're businesslike. Then we give them context. So you are an AI assistant, not a lawyer and cannot provide legal advice. Business hours 8

[1:06:42] to 5:00 p.m. Monday through Friday. Current time we give it this. Then we have the user's phone number. So the user number, these are all variables that we add to the agent. We put this curly bracket curly bracket user number right uh so the user knows also the agent knows sorry consultation length will be 30 minutes we have three lawyers at the firm we have James Mitchell David Chen and Maria Rodriguez we have instructions so don't say this when this happens keep a natural language or keep a natural tone when it comes to names we say first name is Jane spelled J A N E last name is Johnson spelled J O H N S O

[1:07:13] N so all communication style is is basically different rules around how we should speak so part of this is asking questions questions like hey what's your name spell your name spell your email right and so these are all different rules that we give it in terms of saying hey when a person says this then do this when he says um yeah natur use natural filler words such as um and so very sparingly maximum once every two interactions and so this is part of setting guardrails for the agent whenever it speaks to you right uh again you can go a lot more detail than this but I think this is is a good start in

[1:07:44] terms of what you should tell the agent so that we're able to have a good uh good conversation then we have available able tools. So available tools is essentially the ability for the agent to book appointments. And this is where we go to functions and we add the tools here, which we'll go in just a second. But we'll have the tool to book appointments and end call. And so under here, we essentially tell the agent that it can only book the appointment once it qualifies the person itself. There's no qualifying that we don't book the appointment yet. Right? So before booking the appointment, it's going to

[1:08:12] ask you a series of qualifying questions and then once you go through all these questions, then it books the appointment. Right? Then we add rules in terms of availability. So when I ask the agent, hey, when are you available? Where are you free? We uh we set rules on that. And so these are all rules in terms of availability, how it should,

[1:08:26] you know, reply to the agent and so on. And then we go to qualifying questions. So qualifying questions is essentially the different questions that the agent needs to ask the user to get information on before it books the appointment. Okay. So question one, question two, question three and so on. Then we have qualifying criteria. So this person must be a family matter in divorce, child custody, adoption, spousal support, domestic violence, must not be a personal injury, car accident, any matter outside of family law and

[1:08:52] location must be within New York State. So if we say hey uh we are based in Minnesota on this question, then it will say we cannot help you, right? Which is great because that's what a human would do. So in this case, the agent will do. Uh then we have different stages of the conversation which will be greetings, qualifying questions, appointment booking, call closing, and then we have different interactions. So the interactions are important just because of the fact that we want to give it examples of what good looks like. Let's

[1:09:18] say we're training an employee, right? If I say, hey, here's like two different conversations that like you should model these conversations when you have your own conversation, then it will do a much better job because it has an example that it uses, right? So here we have example one, qualified user, successful booking, and the conversation starts [music] with this. Thank you for calling Troy's Law Estate. My name is Grace. I'm a digital assistant here. This call is being recorded for quality purposes. How can I help you today? Well, actually, this should be your main message. So, let me just paste this here. Thanks for calling Totally Law Estate. My name is

[1:09:51] Grace. I'm a digital assistant here. This call has been recorded for quality purposes. How can I help you today? Hi, I need help with the divorce. I'm sorry to hear that you're going through this. What should the lawyers know about your situation? My husband and I have been separated for 3 months and we have two kids. We need to figure out custody and divide our assets. I understand that

[1:10:06] sounds like a challenging situation. What location is this matter in? So you see how it also adds the human aspect of it which is showing empathy to the user. Again the whole point of an agent here is not for the person to think that it's human, right? It's to to speed things up and to like be able to reply to all the customers that answer the phone. And so in adding these it will generate a different flow of conversation and it will not sound very robotic. It will be like a human would uh because a human would say exactly this. And so now we give it the whole like user agent user

[1:10:36] agent user of what the agent says, what the user says and so on. So it can model this when it actually um you know books the appointment and we give it another example as well. So again this right here is the prompt pretty strong prompt because it has different parts to it. The more detailed the prompt the better the output is. So this is part of like the instructions of the AI agent like we mentioned u you know being one of the main parts of the agent is the the the instructions that you give it right if this is bad you cannot expect the agent to do a good job because it's not the

[1:11:04] best right whilst if this is very very detailed step by step with guardrails and rules then it will follow what you say more closely all right so now that we're done with the actual instructions of the agent which is the prompt essentially what we want it to do now we're going to give it the tools so the functions of what it can do so it can take action it can book appointments It can check availability. It can end the call and so on. So this is where we go here. We can zoom in and we can go to functions. So this is to enable your agent with capabilities such as calendar

[1:11:33] bookings, call termination, etc. So we have one here which is end call. If you recall here when I go to available tools, we have the end call. There you go. End call. So the available tools are going to be end call which we already have which by the way is just a prompt. And then now we have to add the uh two different ones. We have to add the check availability and we have to add the calendar booking. So if I go to add, I can see that I have a ton of, you know, I have call transfer, I have SMS, uh,

[1:11:59] IVR and so on. But we have two here. Check availability and book appointments. So we can go in this one first. Uh, check availability. When the user asks for availability, just check the calendar and provide available slots. And now we need the API key for cal.com, the event ID for cal.com and the time zone that we have. So to do that, we can go to cal.com. You can go here. You can actually sign up to cal.com. I already made an account, so I'll use mine. And once you uh sign up to cal.com, you'll get on a page that looks like this. Now, I already have a

[1:12:26] ton of calendar events cuz I use this every single day. But the first thing you want to do is you want to create an event type. So, you want to press new and you can name this lawyer consultation. Uh duration, let's do 30 minutes. For myself, that's fine. Continue. I already have a consultation. It looks like this. So, you can just press continue uh because you will not have one. And that's it. So, we have availability. In this case, mine is from 3:30 p.m. to 12:00 a.m. and then 12:00 a.m. to 1:30 a.m. uh with my time. And right now, I'm in this time zone. So,

[1:12:58] Asia/ Dubai. So, I'll use that. That's it. So, if you are in America/ New York, then in here, you put America/ New York or whatever time zone you're in. In this case, I'm in Asia/ Dubai. So, I'll use this. Now that we made the actual event ID, which is just a link by the way that someone can use to book an appointment like this, super easy. Uh now what we have to do is we have to go back here and on the URL on the top we can get the number. So there's a number between event types and [music] uh question tab name. This is the number

[1:13:29] that we call event type ID and that's the one that we paste here. Event type ID. So it knows which event it's going to use. Now the second thing is that we have to go back to the dashboard and we have to go down to settings to get our API key. So settings uh API key on the

[1:13:45] bottom and we have to make a new key. You can name this retail agent. You can put never expires. That's the one that I put just because I don't want to go back and make a new one every single time. So you can press create. You can go to k.com. You can just copy this API key. Uh and then you can bring it back and paste it here. So now it has the password to your account and it will use this exact event type ID to check the calendar availability. Okay. And that's

[1:14:11] the first one. All right. So that's it. We can press save. And now we are done with the availability. So now the agent is able to check the availability on the calendar. The second thing is uh we have to book the appointment. So now that it knows when we are available, now it has to actually book the calendar event with the user. So we can go to book on the calendar. And then in terms of the description, you can make this more complex, more detailed. I want to keep it very, very simple. We can just leave it as this. So when the user asks to book an appointment, book it on the

[1:14:39] calendar. Yeah. So that's fine. We have the name, description. Now we have the API key. So just copy the API key that we had just now. Paste this here. And then we have to paste the event type ID as well. And then change your time zone. So time zone, just make sure it matches

[1:14:54] the time zone you have here. Concentration one, availability here. Make sure it matches this. And by the way, to set availability, you can just also go here and you can make a new availability. Name this, I don't know, lawyer consultation. Consultation. And then you can set the times when you want to do it. If you don't set this, it's going to do like a default one from 9:00 a.m. to 5:00 p.m. like you see here. Um, but it's fine. It just it just depends on when you're available and when you want to to add your your bookings, right? So we can press save and now we give the AI the tools. So we

[1:15:32] have check availability and we have booked the appointments which is great. So now it has the capabilities of doing whatever we wanted to do. There's one more thing which is the knowledge base. So the knowledge base is going to be the document full of information of Q&As's FAQs um that the agent is going to use to answer questions about the company when the user asks. Now I don't have a knowledge base prepared but we can just make one right now using AI just as a test. I'm going to use a doc here and name this tor law estate knowledge base. I can go to chatbt. I

[1:16:07] can go back to retail. I can paste this here. Uh and then I'll say you are an expert uh knowledge base creator and you create very highdetailed knowledge bases full of Q&As. Now, I'm making an agent, a test agent, and I will give you the prompt of the agent down below. What I want you to do is give me a like a pretty simple knowledge base, a document full of FAQs. You can make up the questions and answers um using this as an example for a video. So, uh give me the knowledge base as well. Now, in theory, you would have an actual knowledge base. You wouldn't just ask to

[1:16:38] do it or the business itself would have like FAQs, but in this case, I don't. So, all right. So, I just finished making the actual knowledge base. I can copy this and I can paste it here. So, it has questions. What types of cases? What cases does the firm not handle? Which dates, attorneys, who handles it, divorce cases, and so on. Amazing. We can uh download this as a PDF. We can go back to retail. We can go uh to add add a new knowledge base. We can press plus here. We can name this toy law state knowledge base. [music] And I can upload a file here. Upload files. And

[1:17:13] there it is. As you can see, you can also add web pages and text. Now, I personally don't like to add web pages for voice agents just because it can tend to be a bit slower, but I just like to get the information from the web page into like a like a document so they can

[1:17:27] answer directly because it is faster. All right, so we added the knowledge base. We can press save. Uh I'm going to delete the one I had before because I don't know why I have it here. It's now vectorizing the knowledge base. Very fancy term of saying it's taking the document and making in a way where AI can read it. Knowledge base. I can press add. I can press this to law estate knowledge base. So now it has access to the knowledge base that it's going to use to be able to answer questions for the user. All right. So now it has all the functionalities that it needs or all

[1:17:55] the different tools and functions knowledge base you know book appointments stuff that we actually wanted to. Um these ones here these settings are mainly specifically to tweak the agent um like how it speaks the voice all that stuff. So the first thing we have here is speech settings and we have background sound. So background sound is essentially the sound of the actual agent. Now typically I do like to use not call center because call center is very very loud. But you can do coffee shop. Yeah, coffee shop I think is one of the best. Very very simple. Response eagerness. I usually leave this uh 0.95. This is how quickly

[1:18:28] the agent starts responding after the user finishes. You don't want it to be like instant. You obviously want to take some time just like a normal person would. Interruption sensitivity. How quickly the agent stops when the user walks over it. This is fine. Reminder message frequency control how often the AI will send the reminder message. This is fine. You can do this as is. [music] Um you can also add some you know words pronunciations in case like you know for example Mikuel some people call it Michelle. So you can basically put the pronunciation of specific words. In this case we don't need to so it's fine. Real

[1:19:01] time transcription settings. Remove noise. Remove back micron noise. That's fine. Transcription is fine. Yeah leave everything as is. It's all good. [music] And then we have call settings. We have voicemail detection. All of these we don't really need. It's not relevant for this. The one thing you want to change is this. So, end call on silence. End the call if user stays silent for extended period of time because we're paying permanent. There are going to be some people that call your agent for fun, you know? So, we want to make this like 45 seconds when it's silent or actually even 30 seconds when they're silenced because no one's going to

[1:19:34] speak, right? Max call duration definitely not 1 hour. You can make this it depends on the business, right? But let's say we make this like 5 minutes and ring duration is fine. You can leave everything else as is. There's no nothing nothing else we have to do. Now we can press um publish and you can press publish. There you go.

[1:19:54] All right. So we built the agent itself. Now it's the time to buy the number and it's time to connect the number to the agent. So we can go back. Let me actually name this so I know exactly which one it is. Tory law estate. And then I can delete that one. You can go to phone numbers, press plus, buy a new number, and you can start the verification. All right. So, when you're done verifying, it should say, "Congratulations, you're done." All right. So, now that we verified our identity, we can go to phone number, buy a new number, and we're going to use Twilio to buy the number right inside of

[1:20:25] retail. So, you can do standard uh we can do in San Francisco purchase. And now we've purchased the number here. Now, bear in mind that the number is $2 per month. So, very, very cheap. But also you would need to change your account billing to pay as you go because obviously you have to pay for this. So you have to add your credit card details and you will pay for the phone number but also for the minutes. Now also bear in mind that they give you $10 for free when you start when you get your account. So for your testing and everything else you won't really have to

[1:20:55] pay for anything. When you work with clients they have their own retail account and they pay for their own retail and you just get access to their account. So it's much easier. Um, so you're not going to have any overhead when you're using the software. So with this said, here's a number that we're

[1:21:07] going to use. Uh, provider Twilio. That's fine. And then we have inbound call agent. [music] And we want to uh we want to connect this. So latest published. There you go. And that's fine. That's it. We also have the outbound agent, which we're going to use later as well, but don't worry about this now. Now, if I call this number, it

[1:21:24] should direct me to [music] the agent. So now we've connected the agent with this number here, right? Super super easy. Um, it took us like 2 minutes. So, it it was really, really seamless. Hence why I like retail AI because it's so so so easy. Now, if we go back to agents, you can see that you have the phone number here. It's connected. So, now let me just call the number and then we can

[1:21:45] talk to the agent to see if it works. And if it doesn't, we'll fix it, too. So, thank you for calling Tory Law Estate. My name [music] is Grace. I'm a digital assistant here. This call is being recorded for quality purposes. How can I help you today? What is it that you guys do? Yeah, Tory Law Estate is a family law

[1:22:03] firm serving people in New York State. The lawyers handle divorce, child custody and visitation, child support, adoption, spousal support, [music] and domestic violence and protective orders. Can I ask how can we who handles who handles the divorce cases? What's what's his name? James Mitchell is the lawyer who handles divorce cases, including property division and spousal support.

[1:22:27] What's his number? How can I help you? James Mitchell's number is 555. 1 2 3 4 5 6 7. How else can I help you today? Uh, I want to book an appointment. Is that possible? Yes, that's possible. Are you looking to speak with someone about a divorce, child custody, or another family law

[1:22:51] issue? Um, yes, I am actually. Okay. What should the lawyers know about your situation? I'm divorced with five kids. Um, and she wants to take all the money. Sorry you're dealing with that, especially with five kids involved. Um, what location or jurisdiction is this matter in? Um, it's in New York. Got it. Thanks for confirming it's in [music] New York. Since you mentioned five kids, there are children involved,

[1:23:22] right? Yes. Yes. Correct. [music] Thanks for confirming. Let me get you scheduled for a 30 minute consultation with one of our attorneys. We have availability on upcoming weekdays between 8 in the morning and 5 in the afternoon. What day and time over the next week works best for you? Can you do Tuesday? Yeah, we can look at Tuesday. What time on Tuesday works best for you between 8 in the morning and 5 in the

[1:23:51] afternoon? 1 p.m. is fine. All right, I'll book you in for Tuesday at 100 p.m. Eastern time. Now, my apologies. It looks like that exact time is not available after all. Let's find another slot on Tuesday. Would earlier around 11:00 in the morning work better or later around 3:00 in the afternoon? Whenever your first slot is on Tuesday,

[1:24:11] I'll take it. Okay, I'll book you for the first available slot on Tuesday at 9 in the morning Eastern time. My apologies. It looks like my system isn't able to lock in a time right now. I can still get you cued for a Okay, so it didn't work. So, perfect because now I can show you how I fix the agent when it doesn't work. There's a few different reasons as to why this happens. So, my first hypothesis as to why this is not working is because I do say here that it's Eastern time and I put America/ New York, but my time zone is Asia/ Dubai. So, first thing I need

[1:24:45] to do is I need to copy this and paste this. So, I can just do slashcurren time uh Asia slash Dubai. There we go. And we can delete the other one. And then Monday to Friday, Asia/Dubai time. So, we'll see if this works. Um I'm not entirely sure that's the issue, but it's all about testing and then see if it works. If it doesn't, we'll we'll check something else. [music] But I'm pretty sure that is the thing that's causing it not to work. So, let's try again. Let me publish this. Uh, let's do 26 June final. Publish. And then when I go back to numbers, I have to update the agent

[1:25:27] there. 26 June final. There we go. Now, let's call the agent again. And let's see what happened. Thank you for calling Tory Law Estate. My name is Grace. I'm a digital assistant here. This call is being recorded for quality purposes. How can I help you today? Um, I want to book a an appointment. Okay, I can help with that. What should

[1:25:50] the lawyers know about your situation? Uh, two kids, divorced. She wants all money. I'm sorry you're dealing with that. What location or jurisdiction is this divorce and custody matter in? New York. Got it. New York. Are there children involved in this matter? Just to confirm. Yes, two. Two children. Thanks for confirming there are two children involved. Let me get you

[1:26:15] scheduled for a 30 minute consultation. [music] What day or time works best for you in the coming week? When does it work for you? Next week, Tuesday. Let me quickly check what times we have open next Tuesday for a 30 minute consultation. We have availability next Tuesday, the 30th, from 8 in the morning to 11:30 in the morning and from 3:00 in

[1:26:34] the afternoon to 4:30 in the afternoon. What time in those ranges works best for you? 3 3 p.m. is fine. All right. 3 in the afternoon next Tuesday, the 30th work. What is your full legal name? And please spell it for me. My legal name is Mikuel. M I C H E L E. Okay, Michelle, I'm booking you for Tuesday the 30th at 3:00 in the

[1:27:01] afternoon for a 30 minute consultation. Now you're all set, Michelle. You're booked for Tuesday. Out. Not Michelle. these agents, man. If there's one thing I hate about these agents is that is that they call me Michelle. There you go. So, here we have the calendar booking U right here. There's one thing I need to change, which is the email, right? Because it says unknown@acample.com, but it got the booking right. So, it it saw the availability, it booked it, and it used the right name. Uh, well, not the right name because it's Mikuel, not Michelle, but aside from that, it booked it correctly. So, now we can tweak the

[1:27:33] prompt to add the email. So I believe we can go here and uh qualifying questions closing ask their full name and I can say ask for their email address as well for booking invitation. And then down here I believe that it also yeah we need to add another thing here. Perfect. Can you spell out your

[1:28:06] email for me? Um, mikuel.t5gmail.com. We can do that. And I can do the exact same thing here. Cuz even in the example, if you don't add that there, it's just not going to ask it. There you go. I believe that should be enough for us to go out and try again and it should work just fine. Final one, publish. By the way, the reason we have to publish every single time is because uh it's like updating a system, right? So, you want to make sure that you update the system [music] over time. And this is what happens as well with clients. When you're deploying this for a business,

[1:28:38] you usually want to listen back to the calls, uh, which you can find here, call history. And you want to be able to optimize the prompt over time. That's part of the monthly retainer that you would charge for this. So, where am I going? Four numbers. Uh, V5 final one. There you go. So, one more time. Let's do it and then we'll see what it cooks

[1:28:58] up. Thanks for calling Tory Law Estate. My name is Grace. I'm a digital assistant here and this call is being recorded for quality purposes. How can we help you today? I want to book an appointment. Okay, I can help with that. What should the lawyers know about your situation? Divorced with three children involved. I'm in New York and she wants all my

[1:29:20] money. Okay, that sounds like a divorce and custody and money issue with three children in New York. What location or county in New York is this matter in? New York City. Got it. New York City. And uh just to confirm, there are three children involved in this case, right? Yes. Thanks for confirming. Since this is a New York divorce with three children, we can book a 30 minute consultation. We have availability during business hours between 8:00 a.m. and 5:00 p.m. Monday through Friday. What day and time work best for you in the coming week? do the first slot that you have on Wednesday,

[1:30:01] next week. Let me quickly check. The first available slot we have next Wednesday for you. Thanks for waiting. The first slot we have next Wednesday is at 8:00 a.m. Does Wednesday the 1st at 8:00 a.m. work for you? Yes. Yes, it does. All right, we'll book you for Wednesday the 1st at 8 a.m. for a 30 minute consultation. What's your full name? And

[1:30:25] please spell your first and last name. James Michael. Thanks. I've got James Michael. What is your email address? And please spell it out. James.mich5gmail.com. Thanks. Let me get that booked for you now. You're all set. James, you're booked for Wednesday the 1st. Thank you. Thank you. Thank you. There you go. So, we booked the appointment. James Michael um five.com. Perfect. So, that right there is a full agent that is able to answer questions from knowledge base, is able to book appointments, and it was super super easy to set up. Now, just a heads up, you can turn this agent into a chat agent as well because I mean

[1:31:05] voice agent and chat agents are kind of the same. They're both conversational agents. The input, which is where the actual user speaks, can be through chat and it can be through voice. It doesn't really matter. But Retail AI, one of the best for voice agents cuz it allows us to be able to make these very realistic, human realistic sound um agents that we're able to use to book appointments, answer questions, and so on. And it's extremely valuable for any business that takes customers over the phone and needs to transfer, needs to cancel, needs to reschedule, needs to book appointments, answer questions, and do all that jazz,

[1:31:34] right? To deploy this again, you would have the number that you would give to the business owner that they can use. this will be here and this is the one that their new customers are going to um be calling to book appointments or if they don't want to use the new number you can just simply connect the number via SIP trunking right and this means that you basically take their number and connect it here to the agent so that some someone calls their number it directly goes to the agent itself so super super easy to set up um again if you are not in the US if you're not in

[1:32:02] in America then I would recommend using telnix or using some sort of other providers and not buying the number Because again this only gives you the ability to buy US and Canada nothing else even tix. If you go here to call history we can see that we uh we can see the logs of the show itself. You can get the recording as well. Uh we can see the call analysis successful the conversation and it's extremely valuable because when you're working with clients

[1:32:28] you'd want to look at the conversations. You want to hear back the conversations so you can see where the agent messed up and where it didn't. So, I recommend also going here and looking at the session outcome and looking at the ones that were unsuccessful and asking yourself why. And then you would have to listen back to those calls to then optimize the prompt and optimize the system over time to minimize the unsuccessful calls and to maximize the successful calls. That's it. All right, team. So, that right there is the second build where we've built a full inbound receptionist, which is a voice agent that can answer questions and it can

[1:32:57] book appointments and it can uh check availability on your calendar. Now that we're done with build two, let's go to the bill three, which is going to be building a speedtole agent. All right, so the next system we're going to build is going to be a speedtole system. So the way that it's going to work is that we're going to have a new lead filling out a form and then getting a call back instantly in under a minute. Now the use case is this. A lot of companies that run Facebook ads, Google ads, any kind of ads, which are advertisements to to basically market their services. people

[1:33:27] opt in the ads, which means that they put their details, whether it's name, phone number, email, whatever that may be. Now, there was a study out there that said that the odds of contacting a lead if called within 5 minutes versus 30 minutes drops by 100 times. [music] And the odds of qualifying a lead if called within 5 minutes versus 30 minutes drops 21 more times. And the reason this is important is because again, this is outbound, meaning that we are calling someone. We're not receiving a call. And so we need to have the speed to lead, the speed to actually call the lead to just, you know, um, give us

[1:33:55] their details to then book an appointment. That's exactly what this system does. So the first step is going to be a type form. Usually this will be like an ad advertisement on Facebook or Google, whatever they may be. Then we're going to use a new software that we haven't used before, which is called NAT. It's an automation platform, super super easy to set up. And then any is going to be allowing us to take the lead submission, take the details, and then transfer it and send all the details to retail AI, which is going to be the software that we're going to use to build the agent to then call the person,

[1:34:23] which will then book an appointment with them and log everything into the Google sheet. So, with that said, let's get to building. All right, so the first step is actually going to Retail AI and making an account. So, first things first, the number that we're going to use to call is going to be uh this number right here. And we're going to connect it to the outbound call agent,

[1:34:39] which is the one that we're making now. Right? Just bear in mind that if you haven't bought the number yet, you can go to plus, you can buy a new number, you can use Twilio or Telnix, um in case you're in the US, you can choose whatever you want. Now, I recommend that for numbers, for example, if you live in Miami or Naples or whatever it is, you get the Naples one. You don't get the one in Pompo Beach. I don't know where that is really, but assuming it's in a different state, you want to get the same number for the same state that you're in. Okay? Because then you have a

[1:35:04] higher pickup rate. People will pick up the phone much better because it's local. Now, with that said, once you bought the number, we can go to agents on the left hand side and we can press create an agent, voice agent, single prompt, built from scratch, and we will start here on this page. Very very similar to the one that we built uh for the inbound receptionist. We have the prompt, we have the settings, we have the first thing that you know the AI will say. Uh we have the different kind of things that we can change and then we have the [music] testing that we can do

[1:35:33] as well. Right? And then we have the number that I can call directly. And in here we can test the agent as well in [music] case we need to. So for the actual um model itself, I'm going to leave this GPT [music] 5.1 or 4.1 default. Simo is fine. Uh English is fine. Then you have the tokens cost latency tokens. That's fine. So I'm going to paste a prompt right here and we can zoom in and we can go through the prompt. So the first thing is the fact that we're still using markdown formatting. Again, this is the way that you know computers understand the different types of hierarchy in terms of

[1:36:06] text. And then we have the role and objective. So you are a friendly solar qualification specialist for GM solutions. You are calling lead name because they recently requested information about solar through a Facebook ad. So you might be asking yourself what is a lead name? So when we are making the type form which is just a software that we make forms, right? The agent is going to reach out to the person and the first thing that they have to say is hey is this lead name is this name is this Mikuel is this James is it John right but how does it know the name and that's when we set these

[1:36:33] dynamic variables so that when the agent calls it receives the the actual variable called the lead name from type form and then it uses it to then know who it's calling right because how is it going to know it's Mike or John for another person right so we use this variable right here and you'll see how we actually transfer the variable to retail AI to say, "Hey, this is the person's name." Call the name or call the number and then say the name. Your goal is to confirm their interest, learn about their home and energy usage, gauge their timeline, and if qualified, book the appointment. Actually, book them in

[1:37:03] for a consultation. Personality, friendly and conversational, instructions, ask one question at a time, keep responses short, track information. If they're not interested, thank them and end the call. Okay. Stages, we have the introduction. Confirm their speaking to lead name. Mention their recent Facebook solar inquiry. Ask if now is a good time to talk. Qualification, are you the homeowner? What's the average monthly electric bill? What what interested you in solar? And are you looking into solar sooner? Just exploring options. This use case is for a solar installation company. So, it's basically companies that have solar solutions like the thing that you put in your roof uh to to make

[1:37:36] things more efficient in terms of energy. And that's what we're selling to these homeowners. I'll come if qualified, thank them and book them in for an appointment. If not interested, thank them for their time and end the call. And then we have the close and important reminders. Perfect. Now we can add the first message. So the first message that we're going to or the actual welcome message, sorry. We're going to have AI speaks first and we're going to do a custom message and then instead of hello, how can I help you today? I'm going to paste this. Hello, am I talking to lead name? Again, same

[1:38:05] exact variable that we have here. And this is pretty much the only variable that we going to give it. So that's fine. But I'm not talking to leave name is the uh name of the person. And then that's pretty much it for the actual settings. So welcome message done. RO and instructions are done. Now we get to the settings here. So I'll leave functions in just a bit cuz we have to just pretty much do the regular book an appointment, check availability, end the call, stuff like that. I'm also going to add a knowledge base for this. So that's that's something I have to do. Uh so

[1:38:35] let's start with that. So let me copy this prompt. Let me go to Chajbt and then say, "Hey, can you make a fake knowledge base for this kind of company? I'm going to give you the prompt of the agent and so on, but it's a solar company uh called JM Solutions." So, here is the prompt that I give the agent. Uh, please make a knowledge base that I can use. Of course, you wouldn't like tell it this to AI because you would have the actual knowledge base from the company with real answers and

[1:39:00] real questions. But here we go. So, Q&A. Do you install the solar panels? Why are you calling me? Is this a sales call? who usually qualifies for a consultant. What do you ask if I'm a homeowner? What do you ask about my lecture bill? These are pretty good to be honest. So, it's questions about, you know, the actual company and then sales sales related questions. Copy this. Make a new doc and then we can paste it here. And then we can say jam solution knowledgebased demo file download PDF. Go back to retail AI and then we can go back here in knowledge base. We can go to plus. We

[1:39:38] can name this JM solutions knowledge base upload files. Choose this one and then we can press open and press save. What it's doing now is basically again vectorizing it making the document in a way where it's very very easy for AI to take the information. Now we can go back to agent. We can go to our agent and in here under knowledge base we can choose GM solutions knowledge base. So very very easy. Uh hence why I like you know retail AI. Then we can do speech settings. So respond how quickly the agent starts responding after you. We can do this. Yeah we can put 0.9. That's

[1:40:14] fine. Uh 0.9 is fine. Speech uh background real transcription settings leave this the same. Call settings. Uh we have to change this. So end the call if the user stays silent for extended period of time. Yeah there are some people who are going to just not speak or they're going to have some problems. So because we're paying per minute we want to pay as less as possible. So if if the call is not really like a real person or whatever it is. So we can put this uh 40 seconds that's enough time for someone to be silent and not say anything. Max call duration depending on

[1:40:44] your qualification but we can do like 4 minutes post call data extraction. That's fine. We can leave this leave this. Leave this. Leave this. Okay cool. So now we've made the the main kind of bulk here. We have to just make the functions. So the functions is the capabilities of the actual agent. So when we receive the type form, it goes to the agent and then what what does it do? Right? So in this case, very similar to the inbound receptionist, we're going to be checking the calendar availability to see if there are slots that are available in our calendar and then we're going to be booking the appointment as

[1:41:14] well. The first thing we have to do is we have to go to cal.com. We have to make an event and then we can we can go from there. Then here uh we have check availability cal.com. When user asks for availability, check the calendar and provide available slots. That's fine. And we need here the API key. And we need the event type ID. Now, for those of you who again who don't really know what cal.com is, it's just a platform

[1:41:35] where you can book appointments, right? It's a booking software. So, the first thing we have to do in cal.com is actually make this event type ID. Um, how do we make this? Well, we have to make an event, right? So, I can go to cal.com, which is this one here, cal.com. I already made an account. It's completely for free, so you're good. And then on the top right, you can press new. We can name this solar

[1:41:55] consultation. Uh, make this 30 minutes and then you can press continue. Set your availability for whenever you want. I have my availability for solar consultation which is 9:00 a.m. to 5:00 p.m. I can press save and then go back to basics. And then on the top right here, that's all you have to do really. So now we have the link that someone could just, you know, directly use to be

[1:42:18] able to book the appointment, right? Instead of, you know, the person doing it manually, the AI agent will actually go inside here and choose the slots, right? So that's fine. Now we can go above on the URL. You see this number right here, 619052. That's the one that we're going to give to our agent here. Event type ID. So now it knows that it has to use that event type with that availability with those available slots and that name of the the actual meeting to be able to book appointments. So if I go back to call.com, I can go here. Now I can get the API key by going down to settings,

[1:42:49] API keys. New speed to lead one. I put one cuz I already have one as you guys can see. Exactly the same. I can put never expires so that it never expires. Never. I don't have to go to cal.com every single time to refresh the API key. I personally don't like to do that. So we can press create and then we get the API key. Save it somewhere, right? Because uh you will not be able to review it again. But you can always make a new one as well. go back to retail AI and we can paste this here. And then we want to change our

[1:43:18] time zone to your time zone. Right now I'm in Asia/Dubai. So that's the one that I'm in right now, but you can just add yours here. So once that's done, please copy the event type ID and copy the cal.com API key. Okay? And your time zone as well because we're going to need this for the literally the exact same thing um for the booking the

[1:43:38] appointment. So save. So now the agent is able to check the calendar availability and then now we have to book on a calendar. So book appointments when user asked to book an appointment book it on the calendar API key we give it here event type ID we can copy this and paste it and then we can change it time zone to my time zone. All right with that said we can press save and now we have the capabilities of the agent itself. So everything past the actual form that the person would fill out, the agent is now able to do bulk appointments and do all

[1:44:14] that stuff, answer questions and so on. So with that said, we can put publish. Press publish here. I'm just thinking if I missed anything, change the name to speed to lead roofing or solar actually and then publish again. Perfect. So now we can go back. We can go to phone numbers. We can connect the phone number to this agent speed to lead solar. And you can do latest uh published. So every single time that we publish it, it would

[1:44:47] just take that one. That's convenient. So now that we're done with retail AI, we can actually take it all the way to step one, which is making the form that the lead is going to fill out before we start making the automation inside of any to connect the form to the agent. So type form make an account and then we [music] can press create form and then we can make a form here. Start from scratch and you can do multi-question page and here you can put uh please fill

[1:45:18] out your personal details. There you go. Full name. I can duplicate this just because I will ask them for their email. Email And then here, I believe it's going to be phone number. Now, I believe that's it really. Unless you want to split up the full name into first name and last name, but I don't

[1:45:41] think that's going to be an issue. Actually, let's do it. Let's make it easy for the agent. First name, last name, and we can name this solar form. Solar lead form. Okay. Again, we're using type form, but the use case would either be actually it would either be a website form for someone requesting a quote or whatever. Uh, but it can also be ads that people run, right? So, they pay for ads and then people would just put their details here saying, "Hey, I'm interested." And then you follow up. All right. So, we just finished the form itself. I believe that's all we have to do here in terms of type form. So, we

[1:46:16] can go now to any 10, which is going to be the place where we'll build automations that look like [music] these, right? So, drag and drop automation. Step one, step two, step three, step four, step five. Nintend is one of the best, probably the best automation platform there is out there in my opinion. You can make an account here, get started. You get a 14-day free trial if you just make an account. Uh, once you made your account, you'll be here. You can press open instance and then you will get to your account that looks like this. Now, yours will be empty, right? Mine is not empty just

[1:46:47] because I've built a ton of automations. But if yours is empty, you can go to create a workflow and then you will get on a page that looks like this. You can either build with AI or we can add the first step manually. I like to do things manually just because I know exactly how

[1:46:59] to build it and what it would look like. So we can go here. Let me rename this to speed to lead uh lead agent solar and then first step will be type form, right? Because this right here is the trigger of the automation like what is the thing that starts the automation and then we can go to step two, step three, step four and whatever it is, right? So step one is receiving the details from the form that someone would fill out and

[1:47:21] then we move on to the next steps. Right? So first step press this look for type form on on the right hand side you can see all the different software that you're able to choose type form trigger again trigger is the first step that initiates the the flow press here and now this right here is your first box right your first box of the automation first thing you want to do is you want to connect your TY form so you want to go here create a new credential and you want the access token to get this go back to Ty form go to forms account top

[1:47:51] right go to account settings and then you will see personal tokens on the left and then you can press generate a new token. Name this speed to lead all scopes because you want access to everything and you can press generate token and now you'll copy the token and make sure to save it somewhere just like cal.com you will not be able to see it again. So save it somewhere. Got it? We can go back to N10 and we can paste our token here. Now, I already had one saved, so that's why I had it there. But you can do Mikuel connection 2 second 2nd of July. Now, bear in mind

[1:48:27] that I typically like to have a naming convention for all my API keys just because uh when you press save, by the way, it will just save it. So, you have this here because of the fact that like API keys will always look like this, you will not really see what account this is connected to. So the more descriptive you are about the actual detail of the thing, the better it is. I always have the same account, but there are times where we use client accounts, you know, and so you want to make sure you say, "Hey, client name connection date, right?" Cuz you also know the date and

[1:48:55] you know who it's for. All right. So now that we connected our account to any 10, we can choose the form that we're going to actually use. So in this case, it will be solar lead form and we can that's it actually. So we can now go inside, we can execute step. What this is is basically saying, "Hey, let's let's just test it." So, we can see the output here. Execute step. Now, what this is doing is it's listening for the test event. It's listening for someone to actually fill out the form. So, I can go here back to my type form. And then I can go to the link and I can start

[1:49:29] filling it out. Mikuel ti. And then I can do john@gmail.com and 20155123 and press submit. So cool, it's finished. If I go here, we can see that now, I don't know if you guys can see it, but it said node executed successfully, which means that the step was actually successful. And you can see we have the output, which is first name, last name, email, and phone number. So, if you got this far, amazing. If yours is not working, then I would check your connection again. I would check your form. And well, it really should be

[1:50:01] working because it was only a few steps. So, it's pretty simple. All right, guys. So, now that we're done with step one, we can two fingers press on this. Rename this uh lead form and then this autosaves. So you're good. The next step now is actually making the call. But in my head, whenever someone fills out a form and they get a call instantly, it's a bit weird, right? As in we want to also emulate human behavior, which means that a human simply will not be there like 24/7 every single second to call the lead. Which is why we want to give it some leeway, some

[1:50:34] time between when the lead fills out the form and when you can call them. And that's where we can press plus to add the next step of the automation. You can press uh you can look for weight. And this wait node says okay whenever we get the details here we can just simply wait this amount of time before we go to the next step. And so here is where we can have a delay of 2 minutes from the time that we receive the details to the time where we call the lead to make it more human. Right now in this case I'm going to use 5 seconds but if you want to do

[1:51:01] two minutes you can do well you can actually just do 120 seconds or you can do minutes and you can do two. I actually don't know the max, but you can do also days like after 365 days, you can just run the automation, right? In case you want to let someone know a year from now, which is a bit weird, but um I

[1:51:17] want to do 5 seconds. Keep it simple. That's it. After time interval, that's fine. You can also do other things as well. All right. Now, um I'm going to press on this. I'm going to press P on my keyboard, or you can just go in here and press pin. [music] What this does is I don't have to re-submit the form on the T form to then trigger the automation. I can just press execute the workflow. We already have the data and I want to test if this works, right? Of course, I knew it was, but always want to double check everything. Cool. This is good. It's all green. Now, we can go

[1:51:50] to the third step, which is actually using retail AI to send the request to say, "Hey, we received this form details. Can you call this a lead and can you do everything you want to do? Book the appointment?" And then we actually go to the Google sheet and add the details there right to the CRM. All right. So if we go here to plus and we look for retail AI there is nothing here. But when I look for time form here there is time for right. So why is it that some softwares are here versus some softwares that aren't right? It's because anytime doesn't have all the

[1:52:20] different softwares but there is a workaround for this and that's going to be using something called API documentation. So if you guys saw the foundation module that I made, I spoke about the API documentation, right? Which is a documentation that tells you exactly how and what you can automate within a software and how to set it up, right? The way for us to actually use retail AI in the actual automation is by going to plus and using something that we call an HTTP request. Right? Don't worry about this for now, but just understand that this right here is going to be the node, the step in the

[1:52:51] automation that we use to say, "Hey, let's send a request to a server. In this case, the server is retail AI saying, here's the my authorization token. Here's my password. Here's the details of the lead. Um, and here's the number that you have to call. Can you just do it?" Right? Instead, of course, of having the retail AI node because we just simply don't have it, right? But if you go to Google Sheet, for example, we do have Google Sheets, right? So if you find that the software is not here, then you have to go and set it up manually, which is actually super super easy. Once

[1:53:21] you do it once or twice, it's just practice and uh it's pretty straightforward. With that said, if I go inside here, we basically have to make this step so that it calls the lead, right? It uses the agent inside retail. So we see different options here. We see the method again, it will get post, put patch, delete like I mentioned before, the URL, authentication, headers, body, and then JSON. Right? So there's actually a way where we can have any do it for us. Let me show you. So I can go to Google and search up retail AI API documentation make a call. If you scroll down, you can see that we have the

[1:53:55] create phone call retail AI. If I press on this, we get to the one and only API documentation. So this is where we will actually put this to practice because you'll see that we'll use this. On the left hand side, you can see all the different actions. create a phone call, get calls, update a call, create chat,

[1:54:10] delete the chat, list chats, and so on. In this case, we really care about creating a phone call. It's a post request because we send the information. We send it. We actually make a call. This right here is the URL that we have to copy. This right here is the actual endpoint that we're calling to the server saying, "Hey server, here's exactly what we want to do." And here's a few things that we have to add. The first thing is the API key from retail AI. Then we have the from number, to number, the agent ID and then we also

[1:54:36] have to give it some examples of data. So lead name, it will be lead name, John Doe, right? So with that said, uh here's basically a description of how to set it up. And on the right hand side, we want to go to curl. So if you're on JavaScript, just change this to curl. And you can copy this curl. And what this is, it's like a bunch of code that allows the server, allows the N10 or the HTTP request to understand what it is that we're looking to do and allows us to send a server request. Right? So, copy this, go back to N10, and then you

[1:55:09] can go inside here, and you want to press import curl, and you want to paste this here, and press import. You see, now everything's pretty much filled up. But there are a few things that we want to change. So, the method post, that's fine. URL. This is fine. We want to get the token from retail AI. So you can just go to retail AI here. You want to go to I believe settings API keys on the left. Add a key API key speed to lead save. And you want to copy this. Go back

[1:55:41] to N10 and you can paste this here. Okay. So bearer space the key. And the reason we do this is because then N10 is able to actually go inside our account. And it's only able to do that if we give it the password, right? So that's fine. And now we have to delete a bunch of stuff here. So put expression go full screen. And now we can delete a bunch of stuff here because we don't actually

[1:56:02] need most of this. Everything from here. So retell lm dynamic variables is something that we do need. So everything from here all the way up to here we can delete. And then everything from customer name in this case it will be lead name. We can delete this as well. So yours should look like this. Okay. From number to number override agent ID and retail lm dynamic variables cuz first thing is the from number. So this is the number

[1:56:33] of the agent that we bought. Two number. This is the number of the person that we have to call. Override agent ID is the ID of the agent that we have to use inside of retail. And then retail llm dynamic variables is the name of the person. So the first thing is me going to retail AAI going to phone number going here and then copying this right here. All this is is just this number but without the parentheses or without the spaces and all that stuff. So copy this we go back to 10 and then we can paste this here. The two number you can delete this and then we can take this

[1:57:06] variable right here and drag it all the way inside here. Again, make sure it's actually in between the the brackets or the the colons. And the agent ID is something that we can get from going to agents. You can press on ID here and you can copy this. Bring it back and then paste it here. So agent this and then lead name will be first name space

[1:57:33] last name. Okay, as you can see now, this is the the thing that the actual agent gets. Mika Tori with a number, agent ID and front number. Front number and agent number should always be fixed. It shouldn't change cuz you always need to call from the same number and you always need to use the same agent. Only these variables change the number that we're calling and the person that we're calling. With that said, let me fill out the form with an actual number that you

[1:57:56] know agent can call and we can start. So, execute workflow. Now, we can go into TY form. We can copy this and paste it. I can go here and put John Tory because why not? We can do my email and then the number. I can put my numbers. All right. Cool. First name, last name, email, and phone number. We can press submit. And now instead of any 10, as you can see, this triggered with this details. It waits 5 seconds. And then it should use retail AI to make a phone call. And it succeeded. So it should call me. As you can see here, I have the phone call from retail AI. can

[1:58:32] answer it. Talking to John Tory. Yes, you are. Thanks, John. You recently requested some info about solar through Facebook. Is now a good time to chat for a couple of minutes? Yes, sure. Great. Thanks. Are you the homeowner at your property? I am the homeowner. Yeah, thanks for confirming that. What's your average monthly electric bill?

[1:58:54] It's uh 20,000. Wow, that's a pretty high bill. What got you interested in looking into solar? Uh, just my wife, you know, she's all about that renewable energy stuff. Got it. Sounds like your wife's really into the idea of going green. Are you both looking to make the switch to solar soon or just exploring options right

[1:59:12] now? Yeah, like ASAP, like next week. Thanks for letting me know you're looking to move quickly. Before we check availability, can I confirm your full name and the best email address to use for the appointment? F. Yeah. Uh my full name is John Thory and my my email is um m i c h e l

[1:59:31] ejmsolutions.com. Thanks John. Just to confirm your full name is John Orte and your email is mckllyjmsolutions.com. Correct. Uh John Torty, not Orte. Thanks for clarifying. So your full name is John Torquy and your email is michellejmssolutions.com. Right? If that's correct, I'll check our availability for a consultation next week. Correct. Correct. I'll check our schedule for next week and let you know what times are

[1:59:58] available for your solar consultation. Here are some available times for next week. Monday through Thursday, 9 to 11:30 a.m. or 3:00 to 5:00 p.m. And Friday, 9 to 11:30 a.m. Is there a day and time that works best for you? Uh, can we do Thursday um at like 400 p.m.? I'll book you in for Thursday at 4 PM for your solar consultation. You're all set for Thursday at 4 PM. The solar specialist will learn about your energy usage, discuss your goals, review your property, and answer any questions. The consultation takes about 30 to 45

[2:00:32] minutes, and there's no obligation. Thanks for your time, John. Have a great day. There you go. That was a that was pretty quick pretty quick build and a very very good outcome. As you can see on the screen, it booked the appointment here. sort of consultation between Minka Torti and John Twerk. I don't know about that, but that obviously can be changed just because Tori is obviously a weird a

[2:00:53] weird name. Um, it's not English name. So, we have additional notes. So, requested ASAP appointment, high electric bill, interested in solar due to spouse interest in renewable energy cuz my wife is all about that, you know. All right, perfect. So, the last step is actually adding the details of the lead to a Google sheet. So, the CRM, right, of the person. So, let's make a Google sheet here. sheets new and we can name this solar inquiries. We can just put full name, phone number, actually no first name. We can do last name. We can do email. We

[2:01:28] can do phone number. Uh I can just go here. I can freeze up to row one. Zoom in a bit more. And then uh I feel like that should be it. Right. So let me go here. Let me press P. Let me pin this as well. Let me pin this as well. It doesn't call me again. And then

[2:01:50] now last step we can go to sheets. We can append a row and a sheet. We can connect our account by going here. You can press sign in with Google. It'll take you to the whole kind of authorization. Very very simple. Once you connect your account, you can go to sheet within document append row which means add a row. And then here we can go to solar inquiries, choose sheet one, and then we can go to lead form because those are the details that we need and we can match these out. So first name,

[2:02:19] last name, email, and phone number. There we go. The thing I like about Nit is that you can see the actual like data points. So you can see whether it's wrong or not, right? For example, if I added phone number here, I can see, oh, this is wrong because it shouldn't be a number. Should be an email. Right? So we

[2:02:35] can just change this to email. Perfect. All right. Now we can test it. So it would just speedrun all of these because it it wouldn't call me as well because it's pinned. And it should just go to the last step. If I go here, I can see that we have John Tory, Jen Solutions, and phone number as well. And the reason why it doesn't do plus one is because of Google Sheets. So this there's a plus, but Google Sheets take that as a it takes it as a mathematical equation, right? What I would also do here is turn this to white and make this like this

[2:03:09] the phone number. Cool. So there there we have the full agent. We have the automation inside of N10 that at the end of the day you'll just unpin all of these. You would refresh. This is the full automation. What you would do is you would simply just publish and now the automation is live and it's going to

[2:03:26] be listening for events um from Tyform. We can go to executions and this is where you'll see the different executions, the different times where the automation run, right? So for this one for example, you can see that everything ran successfully and so on. If this is red, that means something went wrong and you can go inside the actual execution to see what went wrong and you [music] can fix it. Instead of retail AI, this agent is fine. It's

[2:03:48] already published. So you're good to go. As in any single time now someone submits that form, it calls the person. But if I go to call history here, I can see that we have the [music] outbound agent and we can see the status, right? So you can see whether it did well or whether it didn't do well. Perfect. And inside here, I can then see the recording. I can see the whole transcript. This is amazing because this is exactly how you would um listen back to the calls that you do for your

[2:04:17] clients and optimize agent over time. All right, so congrats on finishing the build three, which was a speed tole agent. I'm going to leave again all the prompts and resources in the link down below in my school community. You can check them out. Um and you can pretty much use them to copy paste it into your own agent. All right, now that we're done with build number three, we can go to build number four, which is an AI

[2:04:36] executive assistant. All right, so the way that the system is going to work is that we're going to have you speaking to the actual executive assistant through Telegram. And by the way, Telegram is just a a messaging software, sort of like WhatsApp, Gmail. They're all places where you can text something. You can you can contact someone. And then when we either do a voice message or a text message, it goes directly to the N10 AI agent, which I call the main agent. So whether it's a um you know calendar events, whether it's something to do with contacts or whether it's something to do with emails, it will route the

[2:05:07] message to the appropriate sub aent which we call in this case and then the sub agent will take action on what we want to do. Okay? So it might sound a bit complex and once you see the system, you might even get scared, but it's actually very simple once you break it down step by step. And we're going to

[2:05:21] build it from scratch, so don't worry. Um, so again, you text an AIA agent. It then sends it to each appropriate sub agent and then [music] it takes action on whatever it is that we want to do when it comes to either emails, contacts or calendar. Okay, with that said, let's get into the demo first. All right, so here we have my Telegram agent on the left. We have the input and we have the output. In between the input and the output, as you can see, there's many different steps which we're also going to go through. Uh, it might seem a bit complex because it's many different

[2:05:48] circles and squares and all that stuff. Uh, but it's super super easy once you break it down. So, Telegram agent, let's do a voice message. Uh, and let's say, um, hello. Um, I want to schedule a calendar event with Michael James for tomorrow at 4 p.m. Michael James. And the calendar event has to be called boardroom meeting. And also send them an email letting him know that we have a meeting tomorrow at 4 p.m. So, super super easy. Just uh, yeah, go ahead and do it. As you can see, the actual uh, telegram started. It spoke to the first agent which will contact the main agent

[2:06:20] and then it [music] will use the email and calendar event as well to take action on email take action calendar. Now I get a text message saying that I have sent an email to Michael James. I also create a calendar event. Okay. So let's see. All right. So it just said that it finished everything. So if I go to my calendar I can see that tomorrow Monday 29th boardroom meeting we have it down for 6 and 7 p.m. Perfect. If I go to my email here I can see that we have a message says dear Michael James. This is a reminder that we have a meeting

[2:06:46] scheduled for tomorrow at 4 p.m. Best regards. Your name. So, this right here needs to be changed. Um, so great that we had an error because now we know what to fix. That's it. So, you basically have the ability to talk to the agent itself from Telegram. And we can take a lot of different actions as you can see here on email calendar, you know, contacts, you can even do research, you

[2:07:04] can even create content, all that stuff. And it's all by using your phone on Telegram. So, with that said, let's start fresh and let's build this from scratch. All right. So, we go back here. Same exact thing as we did before. Uh I'm going to paste this here which kind of walks you through the AI agent fundamentals. Um because with NA10 they do have a very specific way of you know building AI agents and the way they look but it all follows the same exact structure that we spoke about before. So we have the input and we have the output. The input is what I tell the

[2:07:29] agent that I want. The output is what the agent tells me that it did. In between the input and the output we have an AI agent which has the instructions again the prompt that we tell it. The AI agent uses a brain. In this case it will be open AI cla whatever it is. It will have a memory to remember the conversation that we're talking to and then [music] using tools calendar, Gmail, contacts, Google Sheets, you know, Perplexity, whatever it is, it will take action on the things that we wanted to do. And we do this by sending a request and getting a response, request, response, request, response,

[2:07:57] right? And so what you saw recently of me doing a calendar event, me sending an email, me getting contacts, they were a series of request and response, request and response. And as you saw, I just gave it a vague kind of prompt and using the AI agent, it thought about what to do and it called the appropriate tool to be able to then um go ahead and and do the actual work. All right. So looking at this from a NA10 point of view, the thing that I showed you before is kind of a representation. What it looks like on the actual NA10 is we have different

[2:08:27] triggers which is Telegram, you know, chat, Gmail, which are basically the way that data comes into the system to the AI agent. Um again in this case that I showed you will be telegram right then we have the AI agent which is something that has a prompt right and it has a way for us to get the input and again I'll walk you through how to build that and then we have the LLM we can choose here memory and then we have the different tools which in this case can be Gmail and calendar event now for the sake of this video I am not going to build a

[2:08:52] simple agent like this we're going to take it one step further and build what I call a multi- aent system and I'm going to go through the multi- aent system because not only do you learn the fundamentals of how AI agents work inside of Nitan, but you learn how AI agents can offload tasks to other sub agents and then the sub agents can then talk to the tools. And this is where we get to the multi-step AI agent where we still have the input, we still have the output, we still have the brain and the memory, we still have the instructions, everything is the same. The only

[2:09:21] difference is that instead of the AI agent talking to Gmail calendar directly, what it does is that it talks to the agents first and then the agents talk to their tools. Now the best way for me to describe this it's sort of like a company right a company has a CEO the CEO doesn't talk to the the VAS doesn't talk to the you know the marketing executive and so on right the CEO talks to the head of marketing it talks to the head of sales it talks to the head of operations and the head of finance right and so this is a CEO these are the head of calendar head of email

[2:09:50] head of blog agent head of contact agent right and so then the head of departments talk to their employees and the tools are the employees of the department itself if I say hey I want to schedule a calendar event it goes to the AI agent. It then goes to the calendar agent say, "Hey, head of calendar, we want to schedule a calendar event." The head of calendar says, "Okay, what tool, what employee do I have to give this task to?" It says, "Okay, the employee to make a calendar event is the one that I have to give the task to." We give a request saying, "Hey, man, we want to

[2:10:15] create a calendar event." And then the response it will tell me is, "Okay, it's all done." So, we review it. It's all good. We then give a response to the CEO saying, "Hey, my employee did it. It's all good." Send it back to the user and the output will be, "Hey, we just finished making the calendar event." Right? And so take that principle for all the different head of departments, head of email, head of blogs, head of content, head of everything. And also one more reason to why we do this is not because we want to over complicate the system, right? Is because the prompt itself can be crazy long. And if the AI

[2:10:45] agent has access to like a million different tools, same with the CEO, there's a good chance that the CEO will mess up which employee he needs to talk to because there's so much context. There's so much information that it has and [music] it will not be accurate and consistent most importantly in the task that it will do. Okay. And so this diagram translates into what you just saw, which is this right here, which is an AI agent that has the input, AI agent, memory, and brain output, and then the different main agent. And then it has the different head of departments, and then it has the

[2:11:11] different tools. So with that said, let's build it from scratch um and get this going. All right. So first thing is to add the input. The input is where the information is coming from. Now we can press first step. We can look for Telegram and we can do okay on message. There we go. So on message and we have this here. So instead of N10 what you see here which is the left middle and right on the left we get the input in the middle we get the we set up the credentials we set up the different settings and the output is the output [music] that the node the specific step

[2:11:39] in the automation will give us okay so first things first we can go here to telegram we can create a new credential and then we need to get the access token so if I go to N10 AI I can ask it how do I get the access token and there's a specific process to go through which is talking to something we call a bot father. So open Telegram and search for botfather. Start with the chat with

[2:12:01] botfather and send the command newbot. Okay. So inside of telegram you can search up bot father. There we go. Press open and you get access to this. You can press create a new bot. Name this I don't know AI executive [music] assistant. Create the bot. Okay. We have to put a username. Executive AI

[2:12:23] assistant. Mikuel bot. All right, cool. [music] And now we have this key. This key is the API token. We can just copy this. We can bring it back. We can go here credential. I can paste it. And then I can say Mikll connection. And then today's date is the 28th of June. There we go. I can press save. Now the reason why I usually name have my naming conventions is because each connection

[2:12:50] can't really see exactly whatever it is. So I want to make sure that I have the exact you know the exact one with a date that I can use. Then next thing is going to be trigger on message. So every single time that I get a message it trigger means starts it sends the information. Um and that's it. I believe we don't need anything else. We can just press execute step. So again we just made the first step right execute step which starts the actual it basically testing the step of the automation. And now I can go here. I go here message and I can start. So it just sends slart. So

[2:13:22] we should get the trigger. As you can see here, we have a bunch of different, you know, we call them data points or variables that we get. We have the message ID from, first name, chat ID, first name, date, text. You see the text was slash start because that was a message that we have here, right? Slart. So we also have a bunch of things here that we don't really care about, which is fine. Um, yeah, that's it. So the

[2:13:45] first step again is getting the message. Now there is one issue or one problem or one thing that I want to add right which is the fact that like which is that sometimes we want to send voice messages right and so text and voice are basically two different methods of communication you know a text to an AI agent and a voice message in AI agent is two different things because a voice has one additional step of using AI to

[2:14:09] extract the text from the voice message. Okay. So what we do here is if I go inside you can see that um where is it? It's it should say text somewhere. There you go. So text if I know that this variable exists it means that we send a text message. But right now for example if I go to execute step and I go back here hello hello welcome welcome welcome. I go here you can see that now we have voice. We don't have text anymore. Right? And so we basically have to set up a what I call conditional logic routting where we say hey send it

[2:14:41] this way if it's voice send it that way if it's text. Okay. So we're going to go here. We can go plus we can look for data transformation or actually not flow. And we can look for if. So if this is here then send it this way. If not then the other way. So if the voice here exists string exists wrong type object. Okay. So if a variable inside here, so let's say if a variable, which in this case it can be any of these, right? I'm basically saying if this exists, which it wouldn't for the actual

[2:15:14] text, then we can um then we can do it. So if this exists, then it's true. If not, that is false. Okay, so I'm going to go here and press P to pin the data. So I don't have to redo it again. And I can execute the workflow. And now it will go to true because this variable exists. Okay, meme type mine type, I don't know how to say it. It exists. Now if I unpinit this and I press execute workflow and I go back to telegram and I just say hello it should go down to false because it's text right and so very very good for us because now we can

[2:15:47] route the information directly let me execute workflow and I can go back and I can say hello hello hello hello hello all right there we go it's true so we can go back here we can look for telegram node once again which is get a file and now we can go down to file ID And we can drag this across. We can download. Leave everything normal. It's file get get the file ID because now when we get the file, we can give it to let's say chat GBT or OpenAI to take the text from that file which is an audio

[2:16:18] file and essentially give us the text. So I'm going to go here and I'm going to use um chat GBT open here and I'm going to transcribe a recording. Transcribe recording. [music] So now the first thing you want to do is you want to go and get your API key from OpenAI. So if you go here to create a new credential, you need to get an API key. So we can do

[2:16:40] this by going to platform.openai.com. Sign into your account. And then you want to go to API keys on the left. You want to create a key here. Name this AI assistant. And then leave everything the same. So all permissions you and then you can copy the API key that you have here and paste it here. And again I'll do Mikuel connection 28th of June. Save. All right. So we have audio transcriber recording. We

[2:17:08] have data. And we can leave this as is. That's fine. So don't worry about anything else. You can put data because that is going to be the data point that we're going to get. So now let's just let me pin this. And let me execute the workflow. And now it should download this. You can see that this is data. If

[2:17:24] I press view, I can Hello. Hello. Hello. I can hear the voice message. If I go here, it should take the data, right? And then now we can see that the text is hello, hello, hello. Okay, so this is the process of transcribing an audio file into text. [music] Okay, and by the way, this is what we call data transformation. Getting data and then you know we call it parsing it, changing it in a way where we wanted to or getting data in different formats and then setting it different ways depending on the format and finally, you know, bring it all back together. So now what

[2:17:54] we can do is I believe there's one more step that we have to do here, but I'm going to I'm going to wait until then. I'm going to make a sticky note first and call this this color is horrendous. Let me do It's even worse now. Let me do blue. Even worse now. Okay, we can leave this this brown and take this out. But I do this because then you guys can visually see where everything is going. So now we can go to the AI agent step. So this is the input. We get the message. So two

[2:18:27] fingers you can press on this rename. Get message. And now we can press plus. We can look for AI. And then we can look for AI agent. So generates an action plan and executed. Can use external tools. Perfect. So let's go off for a second. And this is what the AI agent will look like. Okay. So I'm going to connect this as well to this because again either way either text or voice it needs to go to the AI agent. And now this is where we get to building the actual agent. Okay. So the agent has a prompt inside that we have to build a chat model which is the brain the LLM a

[2:19:00] memory tool that we're going to set up and then tools which are going to be first of all sub agents and then the sub agents will have their own tools. Okay. So the first thing I want to do is I want to go inside and make sure that it knows that the input that it receives is the one from this. So if I go here, I can change this from connected track to your node. I can replace this with the find below. And the prompt which is a user message again will be tell the AI agent that we want to do will be text I want to get my contacts. Let's set up

[2:19:29] the AI agent first and then we can we can see if there is an issue to solve. But okay, so this is the user message which is the thing that we tell it to do. Sort of like the prompt you know when you guys go to trajet and you say I want you to make me an email whatever that's a user message. Okay. Now there is another prompt because user prompt is what I tell it to do. And then we have instructions. So the rules and you know identity of what the agent needs to do, right? So we can go here to add option

[2:19:50] and we can look for system message. Don't worry about any of these. We really don't need them. The most important one is this. So the system message is essentially a prompt that gives AI the ability to understand what its role and identity is and what actions it needs to take. Right? So I'm going to paste the prompt that I have here that I already made and we can go through it. So first things first, you can see that we have markdown formatting. Markdown, markdown, markdown, right? In a way that, you know, AI knows the hierarchy of what matters. And the first thing is the overview. So we tell it that it's an

[2:20:22] ultimate personal assistant. The job is to send the user query to the correct tool. That's the main role, right? As a CEO, your job is not to send the emails. It's to send the task to the correct head of department. And the head of department can then worry about the actual thing itself. So then we have tools. Now the tools is essentially the thing that the main agent or the this agent has access to that it can send the

[2:20:42] information for the action to be taken. So we have contact agent, we have email agent and I believe we have calendar agent. There we go. So contact agent uses tool to take contact actions. [music] You must use this before sending emails or creating calendar events within attendees. So essentially we're going to feed you know this agent with our list of contacts that we have. Sort of like a CRM but it's going to be on Google Sheets, right? The issue is that if we tell AI agent that we want to, you know, schedule a calendar event or send an email to someone, it first needs to know who the person is and then it can

[2:21:13] do the action, right? Hence why we tell it that you must use this always before sending emails or doing calendar events. The email agent, you know, use this tool to take action in email. Calendar agent use this tool to take action in calendar. You must use the contact agent before using the calendar agent and make sure you pass through the email of the contact to the calendar agent. And so these rules are extremely important because they allow the AI to have those guard rails, right? Then we have more rules. So some actions require uh you to look up contact information first. For the following actions, you must get

[2:21:41] contact information and send that to the agent who needs it. Sending emails, drafting emails, creating calendar event with attendee. Instruction: Call the necessary tool based on the user's request. Use the think tool to verify uh you took the right steps. This tool should be called every time. Final reminder, here is the current date and time. And then we give it this. Now this right here is just a variable. Like super super easy. First of all, the reason why we tell the AI what date and time it is is because it is not the best, right? So to do this we can do curly bracket curly bracket. We do now

[2:22:12] and you can pretty much leave it at that. The one we had before which is this. It's just for the format, right? But we don't really like the AI doesn't really need the format. It just needs the date of today. If you want to format it, right? In case you're interested, you can do dot. So again we have bracket bracket and inside here is the actual thing and you can see the output that we have here dot format there we go converts the date time to string format and then we have different options so we can choose year year month as you can see here now it's 2026 06 28

[2:22:52] so this is fine it's good enough either one is fine you can also just leave it as now as long as knows what date and time it is today. Um, it's going to be smart enough to convert the actual, you know, the the the date format. So, cool. This is the prompt that we're going to use and that's how it's going to think or the basically the context it has

[2:23:07] behind every single thing that we do. So, now that the agent knows exactly what it needs to do, the identity, the rules, the tools, we have to actually give it the things, right? So, the first thing is the chat model, which is the brain. Now, the brain, we can go here, press plus, and you have access to different LLMs. Right? Now, I like using OpenAI because it's pretty easy and also it's pretty good. So, if you connected your OpenAI connection already you which you did before, you already have it here. We can go to 5.4 mini. There we go. Use responses API that's fine and we

[2:23:39] can leave this as this. Okay. So, now we can rename this brain. Again, you can do this two fingers press on this and you can choose rename. Next thing is the memory. So, we press here. You can do simple memory. You can also do some complex memory. I did make a video on all the different memory nodes and

[2:23:54] memory different things in my channel. Um you can just go on the channel and search up memory nodes and A10 mikuel and you'll find it. Um but we're going to use the simple memory here to keep it super simple. So the session ID is actually something that we need to change. So the reason that is is because there's two different ways of um like chatting to the agent. One of course is through telegram but we also have a native uh like node inside native step inside of any which is a chat trigger node which is a node that we can obviously chat but it's it's any native

[2:24:25] right meaning any native so and each session that you talk to the agent has a session ID right now for telegram you can press the find below you can find it here so get message chat ID there we go you can drag it across and it'll be here just a number right and now it's going to ask us how many past interactions does the model receive as context. So how many messages will it have as context that it's going to remember before it gives us the answer, right? So if I say, hey, send an email to John. It sends an email to John. And then also I

[2:24:55] can say schedule a calendar event with him. And it can do that because it remembers who John is because it has memory, right? So I can leave this as 10 to keep it super simple. Um I think five is a bit too short. Now there's one more thing that I want to add here which is

[2:25:08] called the think. The think step, right? The think step is something we can we can pull here from tools. You can search up think tool. What it is is this, right? Use a tool to think about something. It will not obtain new information or change the database, but just append [music] the thought to the log. Use it when complex reasoning or some caching memory is needed. And the reason we do this is because when we have sub agents and we have, you know, the main agent and different tools and calendar, email, all that all that jazz, it is important that it makes the right decision. Now, it is going to do a good

[2:25:37] job here with both of these, but it does even better if it has this second opinion, right? Sort of like a surgeon having a second opinion in whatever he does, right? This right here is a second opinion, and it is going to be a tool uh that we're going to use. All right. So, now that we did this, we're going to put this to the side. This might get a bit confusing in terms of how it looks, but trust me on this. You want to make sure that it's very organized because we're going to have a whole a whole set of sub aents here. So, I want to do the output

[2:26:00] first, and then we can do the actual sub aents. All right. All right. So, I just deleted the actual box itself cuz I thought it didn't make sense to have it there. But what have we done so far? We have made the input. So, we get the message whether it's in a voice format, whether it's a text format. You know, we transcribe the recording. If it's a

[2:26:13] voice format, if it's text, it's text. Then we give it to the AI agent. We made the system prompt and we made the user prompt. Then we connected the chat model which is the brain. Then we made the memory because now it remembers the conversation. And finally, we made a tool, not we didn't make it, but like we added a tool, uh, which is called the think tool, which allows the agent to have a second opinion behind every single decision that it does, which is great, especially when you're building something a bit more complex, which we are. Not complex, but it requires different steps. Right now, we want to

[2:26:40] do the output. So, I can go here, press plus, telegram, and then send a message. There we go. Send text message, credential message, send a message. We can do the chat ID which we can get here. Get message chat ID. Again, each conversation is a chat ID. Think of it as like you know each time you have a conversation with your friend that is an conversation ID. If it's a different conversation, it will be a different ID. That's important because that way it gives me back the messages in the same exact you know conversation. It doesn't send it to like another place. Let's run this first. Let

[2:27:12] me just put hello and let's run the agent. I can go here. I can say send an email to Michael James. And now the AI agent will think there's a problem. We'll fix that problem. But as you can see, the data went through all the way here. So what was the problem? If I go here to simple memory, provide a key to use a session ID in the key parameter or use the connected chat trigger node option to use a session ID chat trigger. So let's try actually to use this. Although I'm not too sure if that will work. I'm kind of figuring it out as you guys. I

[2:27:44] typically use the session ID of the Telegram chat, not the one that I'm having with the agent. But let's try. Let's try. I'm going to pin this. Just cheat code in case you don't want to do it all over again. I'm going [music] to press execute workflow. And there we go. Another problem. Expected

[2:27:59] to find a session ID in an input field. Call session ID. Okay, that Yeah. So, it's not going to so I'm going to do a key. Get message. Yeah, I would assume it's this chat ID because it's green. Let's try it again. Okay, now it works or it doesn't. Never mind. Let's go back. Okay, I think I think the issue might be that we cannot add a variable, but I want to do I'm going to copy this and replace it with this because the chat ID is always going to be this, right? It's always going to be the same kind of conversation with the agent. So, let's try again and let's

[2:28:32] see if it works. Okay, there we go. Now, it works. Cool. Now, we have this here. Memory is fine. Uh, we can connect this to the agent. I assume that I'll also have to change this to the number and again the number uh it will be different for you guys than it is for me because my conversation ID or my chat ID is different right you can get it here chat ID just copy this and you can paste it here now for the text I can go to the AI agent and I can get the output right JSON output and you can see a pattern here every single variable that

[2:29:03] we have when I drag this across it's all code right we don't really have to care that it's code and so if you're looking at this and you don't really know exactly what is going on And just know that this is a dynamic variable, right? It's the variable that changes every time. And so this is the output that we get here. Output. And here it will look like this. So we send the data to one step to another. All right. So with that

[2:29:24] said, let's go execute the workflow. Memory is fine. And if I go here, I still need the email message to send. Please send the body of the email. And if relevant, which James Michael you mean? Now we also get this right here. This message was sent automatically with N10. I don't want this. So to delete this, we can go to add field. We can go to append and attribution. We can turn this off. And if I execute step, I get the same message without this. So quick one here. All right. Now that we're done with the boring stuff, let's try let's try, let's make the agent. If I go

[2:29:54] inside, we mentioned that we would have the contact agent, which is one, email agent, and calendar agent. So let me just copy this. So I'm going to go here outside. I'm going to press add to tools. And then I'm going to do uh sub or agent. There you go. So, first agent, second agent,

[2:30:19] and third agent. This looks sick. It's awesome. Okay, three agents. So, we have to make sure that the names of the agent are the exact same as the names that I put in the prompt of the main agent, right? So the first one was contact agent, email agent, and calendar agent. Let me actually put this. Let me make sure that this right here is like this. C small and then a big e small and

[2:30:47] then a big. There we go. Perfect. Personally, that's what I like to do. Um, you don't have to do it that way, but you can if you want to be like me. Okay. Calendar agent. Uh, contacts. First, we do contacts agent. Then we do email agent and then we do calendar agent. There we go. So now we have to do the exact same thing that we did here, we do here. We have to make sure that we set this up with a model a

[2:31:17] memory and tools. There is a quick hack. Okay, the hack is that instead of going here and, you know, choosing OpenAI again and choosing the model and whatever and then having it here and doing that for each one, we can actually connect the chat model to the one that we always had like this. There we go. Now, this right here, you can do even if you don't do it, it doesn't really make much of a difference just for speed sake. I know in terms of visuals, it's looking like the matrix here with all these different lines and stuff. So if it looks a bit confusing to you, then

[2:31:46] you can just like take these out, you know, and kind of have your own chat models here. I would just copy this, paste it, right? I personally do this because it's easier for me. So [music] it's all good. Now that we're done with the actual brain, we don't actually need the memory. So the memory is not something that we need because the memory is something that the main AI agent needs, but not the head of departments. The head of departments just have [music] to execute on a task and that's it. Right? So this right here, don't worry about it. Now we have to set up the tools which is contacts,

[2:32:14] emails and calendar. So the first thing we have to do now is set up the prompts for the different agents. So if I go inside here, I see that this is the exact same as the one that I had before. The only difference is that we have a description. So for the description, I'm going to say this call this tool to take action within the contact database. And then the user message I can press this button which means that the model the AI inside of any 10ent for the agent [music] is going to define that automatically. So the input that's going to come in is going to be defined

[2:32:43] automatically. And now I can add an option to put a system message and I can go here go up and I can paste this message right here. So super super easy very similar to the one that we had before but you know very much simpler. So we have overview you are a contact manager. You have tools to take action within the contact database. [music] The first tool you have access to is get contacts. The second one is going to be add or update a contact. Use this to get contact information like email, address, or phone number. Use this to add a new contact to the database or update an

[2:33:14] existing contact. Perfect. It's all we have to do for the prompt. And now we can start adding the tool uh itself. Now I made a [music] database on Google Sheets which looks like this. You can pretty much make the same one. It doesn't matter. But this is going to be the contact database that we're going to have. Now typically a company wouldn't really have Google Sheets. they would have an actual CRM like HubSpot, you know, so you can use that. But in this case, I'm going to keep it simple and use this. So once we have that, I can go plus the tools. I can go to sheets,

[2:33:41] Google sheet tool, [music] and I can make sure that this right here is my connection, which I can do by going here, sign in with Google, and then choose your account. Once you've connected to your account, set automatically she within a document. You can get rows because this is the first action to get contacts. Uh the list will be contact database USB because that is the name of my Google sheet. The sheet name is going to be sheet one. And that's it I believe. There we go. And now I have to name this the name of the actual tool that I told it. So get contact, add or update a contact. There

[2:34:15] we go. So get contact. Okay. I'm assuming you're starting to get the gist of the whole process, right? you have here agent sub agent and then the agent has tools and then we we pretty much do the same thing for all these right so that's the first tool there's nothing else you have to do apart from choosing the sheet it will do everything automatically so don't worry the next thing is we have to make the add or update a contact so I can go here plus sheet Google sheet same connection everything is the same apart from this append or updated row choose a document which is the one that we had before

[2:34:49] sheet is going to be sheet one column name to match on will be name and then here. So this is saying that if we have to update a row, we have to know what row it is, right? So what variable, what thing do we match the the contact with, we can match it with this and then we can update whatever, right? So I can go here match with name and then in case we have to append, which means add a row, add a contact, we can let AI, you know, kind of define the name and we can say description, this is the name of the

[2:35:18] person. We can do the same with email. And I can say this is the email of the person. Now a lot of AI is used here because we want to make it as easy as possible. And so the AI is here. If we have to add a contact because it's an agent and it's talking to us back and forth, it will know what name and [music] email we need

[2:35:38] to send to the Google sheet, right? Because it's going to ask us. So we're going to use the AI instead of N10. We're saying, "Hey, agent, figure it out." like use AI, use your brain to think of what name and email we have to send through based on what the user is telling you in the conversation. And now I can do append or update

[2:35:55] contact. I believe that's it. Is that the name? Or add or update contact. There you go. Cool. So let's start. Let me actually test this so you guys can see uh what this looks like. Let me unpin this. Execute workflow. Let's say I ask it to give me the email for Michael James. Give me the email for Michael James. It will talk to the agent

[2:36:22] here. It'll talk to this. Bring it back. Bring it back. And then send it here. The email for Michael James is nic. System77.AI. As you can see, it's accurate. So, very, very good stuff. God, this looks so complex with all these. This looks crazy, but you can see it's much easier than than it looks, right? Um, so it's fine. All right, that's it for the context agent. Now we can go to the email agent and the calendar agent. So pretty much the same exact process. We have to go here inside. We have to put the description. So call this tool [music] to take action email the prompt

[2:36:57] user message. We can press this button which means that we're letting the model define this you know by AI. And then we have to make a system message. I can go here expression. The reason why I'm doing expression is because I can't go full screen. And I can paste the prompt right here. So you are an email assistant. You have the tools to take action email. The first tool is send an

[2:37:15] email. Send. Second tool is get emails. Then it's email reply. Reply. There you go. Create a draft. Get labels. Label emails and so on. There's a lot of tools, but it's super super easy to set up because it's it's just repetitive. You just have to, you know, repeat the whole process with each single tool. So, I can go out and also we're telling it the date and time because when we're sending emails and we're saying, "Hey, send an email to John saying that we have a meeting tomorrow." It needs to

[2:37:38] know what today is for tomorrow. Cool. Nothing else here. I can go tool. I can look for Gmail. Make sure it's not human in the loop. That's a different thing. We can do Gmail tool. Make sure it's a Gmail tool. And make sure to connect your email by going here, sign in with Google, and then you'll press your account and log in. Once you have your email set up, you can go to set automatically message. In this case, this first action will be send email. We're going to press AI AI AI because we want AI to define the actual um variable here. And we can say

[2:38:13] this is the email address of the person we want to send the email to. We don't need the description for the subject line cuz it needs it knows exactly what to do. And message it will be the body of the email. So AI can decide this. That's the first one. So we can name

[2:38:28] this send message. Then we can copy paste one here, one here, here, here, and here. Uh because now I can go into each one and just change the action, right? So instead of message send, we have to get emails. So message and we can get many basically get all the emails. I can press this. Let AI do it. We can turn this off because we want all the information from the email. If it's if you use simplify, it will just give you like a very very few data points from the email and sometimes we want more stuff. We don't want to be restricted in what we get from the

[2:39:04] emails. Hence why it's off. That's it. So get emails. Get email. The next one is going to be email reply. So if I go here, I can go to message reply. Message ID. I can use AI. And then message. I can use AI as well. There we go. Perfect. And I can name this reply email or email reply.

[2:39:27] Actually, no. Reply email. I mean we do have to make sure that the names are same same here right as send email. Then we have the last three which are the first one is going to be create draft. Then we have get labels and we have label email. So create a draft very straightforward.

[2:39:52] You're just creating a draft for email. But for label emails to be able to label an email, we have to get the labels first and then get the ID of the label and then label it. Okay? Hence why we have two different steps. So before it uses this, it always uses this. Create a draft message. Um is it message or draft? Draft. There you go. Create a

[2:40:13] draft. AI defines this. AI defines this. It's all good. Just make sure this is HTML because we already told the agent here that it needs to make the emails in HTML which is a very very specific type of way of writing things for AI. Get labels. We have label get many return all. We just put AI. Basically, we're letting AI decide everything. Connect this here. Connect this here. Connect this here. And then finally, label emails. We can do message and we can do add a label. Message ID here. And this will be here. And here I can put this is the label name

[2:40:53] or ID. Cool. All right. And now I can connect this to this. So we have just finished the email agent. So let's try this out. XQ workflow. I can say um send an email to Michael James saying that we have dinner tomorrow. Send an email to Michael James saying that we have a meeting tomorrow at 5:00 p.m. There we

[2:41:15] go. transcribing. There is a problem. Let's see what the problem is. The problem is that the text message is not it didn't come through. But if I go here, it should come through. Okay, I think I know why. That was a thing that I mentioned that we might be missing. But a very easy step here is I can just new line here and I can just send the

[2:41:40] text here. Execute step. See if it works. There we go. Email agent, send an email. We have a meeting tomorrow at 5 p.m. Best your name. Okay. So, [music] we had this problem. So, let's fix this problem first. And then let's also take out this email was sent automatically with any. Um, one quick one. The thing I just did here is that like when we get the text message from Telegram directly and when we transcribe the voice message, as you can see, there are two different types of variables. One has dot message and one doesn't have dot message. So, I'm going to add both variables in case, you

[2:42:12] know, the voice message comes, it takes a voice message, the text message comes, it takes a text message. It's pretty easy. So, we have to tell the agent that it needs to use our name. Please replace or use the name instead of your name. at the end of the email. Always, always, always. I don't know why I always use please, as if an AI agent is going to get mad at me. And then I'm also going to add an option. I want to remove this, right? It's called an Nintend attribution. I believe I can do

[2:42:59] it here. Can I do it? No, we can't. So, I don't think we can actually add an option. Oh, we can. Okay, there we go. Append attribution. Turn this off. And reply email as well. Turn this off. And then create a draft. Yeah, I mean it wouldn't be there in the first place, so it's fine. Cool. Amazing. So, we just finished with the email agent. Now, let's go to the calendar agent. So, let's go inside and let's go to the actual description. So, call this tool to take action in the calendar. Then, the prompt will be, you know, AI. And then we can make the system message,

[2:43:35] which really simple. You can go here full screen. Basically the same thing as the Gmail agent. We say you are a calendar assistant. You have tools to take action in the calendar. Create event. Create event with attendee. Get events. Delete events. Update events. Perfect. Uh memory. We don't need. We can go to calendar Google calendar tool. I'm just going to copy paste. I believe it's five tools. I think so. So here first thing we have to do is connect our Google calendar. You can go here, sign in with Google. Same exact process as Gmail. The first one was create an event. So create event. You can change the name also

[2:44:15] here. Event create from my calendar list. So you can do my email. And then I can let AI define this and define this. Second one is a create an event with attendee. So basically the same thing. Choose AI. Same with this. But now we want to add an attendee because sometimes I want to create my own calendar event and sometimes I want to

[2:44:37] create an event inviting someone else. And here I can define this. This is the email of the person we are inviting for the event or we are sending or we are sending the calendar the calendar invite to. All right. So actually let me just change this. Create an event with assenee attendee. I don't

[2:45:04] know why I call it assigne. There we go. Next one here is get events. So create instead of create we do get many calendar. Choose your calendar. Let the model def find this, this, and this. Super easy. Get events. Then we have delete events and update events. I should copy this here. I can go here. Events. delete calendar event ID use AI delete events and again the reason why we add names is because we told it hey use this tool when this happens use that tool and so if this is delete events but it's actually creating an event it would call the wrong tool and so each name

[2:45:46] corresponds to the agent itself and here finally we have update events so this would be update ID there you Awesome. So, I think we just finished. Yeah. Let's connect all of these. I don't know why it keeps tripping. There we go. So, as you can see, this is the original agent that we had. Right. So, now I can press execute workflow and I can say send a calendar event to Michael James uh for tomorrow at 5:00

[2:46:16] p.m. and call it boardroom meeting. So, you can see here it called the contact agent first and then it called the calendar agent and then it finished. So, done. I created the event boardroom meeting, okay, for tomorrow at 5 p.m. and invited Michael James. System 7.ai. Perfect. Cool. If I go to my calendar, I should see that I have boardroom meeting here with the email itself. And by the way, as you can see, it called the create an event with attendee. Right now, let's say I want to update the event and I want to shift it for Tuesday at the same time. Can you shift the

[2:46:55] event so that it's the same time but Wednesday actually no Friday. There we go. Now it will get the event. It says I can help with that but I don't have the tool here to edit an existing calendar event. If you want to create a new event. Okay. So interesting. It says it doesn't have access to it but it does.

[2:47:15] Maybe it's something within the prompt. If you take action calendar, you must use account take action calendar to update events. And what else is there? We have to update events and delete events. There we go. Now, if I execute workflow, if I give it the exact same message, it should send it to the calendar agent. There we go. It should get events first and it should update the event. And now we should see the event on Friday. It says, "I tried my calendar. This tool

[2:47:45] didn't reply." It says it didn't work. Okay. I believe it didn't do it because I didn't specify what event. So, this might be a fault here. Yeah, I can just say, okay, I have an event tomorrow which is called boardroom meeting. Can you update it so that it's on Friday, not tomorrow? Let me just execute the

[2:47:59] workflow. Okay, I can just press send. As you can see, it triggered the AI agent. Now, it's going to use the calendar tool, calendar [music] agent to get the event and then update the event here. And then it's sending us back a message on Telegram. But if I go to my calendar event, I can see that if I refresh, it moved the event from Monday to Friday. All right, so that right there was a full AI executive build. As you see here, it is a bit more complex than the ones we've built before. Maybe not super complex, but it requires different steps and more features and

[2:48:28] more stuff. Now, this is really, really powerful because it allows executive CEOs to be able to tap into their emails, tap into their calendar events, tap into their contacts as easy as just talking to to Telegram, right? Sort of like your personal assistant, but in this case, it's an AI executive assistant, right? That basically does all the things for you. Now obviously Gmail calendar and context are the first use case. But we also have other things depending on the person. Maybe someone might want to take action in their own CRM. Maybe someone might want to create content and post content on LinkedIn or YouTube or what Instagram, right? And so

[2:48:59] you can do that by adding more and more agents. Just make sure that if you add more and more agents, you obviously don't overload the prompt too much on the main agent because then you can start to make mistakes um in where you know things are at. All right, so congrats on finishing build four, which was the AI executive assistant. Now, we're going to go to our last build, which is build five, and that will be an

[2:49:19] AI sales lead research agent. All right. Now, the way that the sales lead research co-pilot works is that we have every new lead that gets researched and written before your rep calls them. So, a lot of sales rep, people who call leads to sell them something, typically want to have as much information about the person before they call them. So, then they have more context. so that you create more professionalism because you

[2:49:43] have more information about the buyer. The less information about the buyer, the more questions you have to ask on the call and the less prepared you look on the call for some offers, of course. So, the way that it's going to work is that we're going to have a type form, which is going to be the same exact um software that we use uh to build the form that we fill out, which in this case is where the person would fill out, let's say, on the website or ads or even to book a call. anything where the lead can fill out the details and that can be

[2:50:09] name, LinkedIn profile, website and email. Then we have clock code which is going to be the software that we're going to use to build this agentic workflow. Okay, so this is different from the agents that we built before. Um, and I'll explain why, but claw code is going to be where we're going to build the agentic workflow and what it's going to do is that it's going to go to the LinkedIn profile, scrape the LinkedIn profile, like get all the information. is going to go to the company website, take the company website and then what it's going to do is that using AI it's going to create a

[2:50:37] brief and what a brief is is a information about the person that you know we are getting information from from their LinkedIn profile and website that we can then put in a Google sheet which then the sales rep can use and can you know can call the actual person itself. Okay, so type form it's research and then add to the CRM in this case it will be Google sheet just because we want to make it as easy as possible. So, with that said, let's start with installing Claw Code. Now, in terms of pricing, we are actually going to be paying for Claude and through Claude, we

[2:51:07] get Claude code. It's sort of like a bonus. For example, you pay for Amazon Prime and you get audible and you get all the stuff that comes with it, right? And so, in this case, we pay for Claude and we get Claude code. For the free version, you can't access Cloud Code. You either need the pro version or the max version. I use the max because I use it quite a lot. Um, but a pro version is going to be more than fine for you starting up. in terms of hosting which

[2:51:26] is where we're going to use cloud code. We can either use it on VS code, cursor, anti-gravity, terminal and claude itself. Now this three is what we call IDE integrated development environment which is a place where very fancy name I know but it's it's basically a place where you can install claw code and you host cl code there you basically use it as an extension and it's very very easy to to set up. The terminal is also a place where we can use cl code. I'm not going to talk about it in this video just because [music] it is a bit more complex for beginners starting out and

[2:51:56] you can also use claude which is the simplest um to look at. You can basically do less things than somewhere like these three. Okay, so in this case you can use VS Code or anti-gravity, right? Pretty much the same thing. I personally like VS Code because it's super super easy to install and you can get going really straight away. So the first step is going to

[2:52:15] code.visisualstudio.com. You want to download for Mac OS. And once you download it, you will get on this page. It looks like this. You want to go down to the extensions. Look for claw code. I already have it installed as you can see here. But you look for claw code. Make sure it's the anthropic one. Okay. And then you can press install here. So if I uninstall and install again, you'll see here. And once you install it, it will take a few seconds. And you should see a little

[2:52:40] logo from Claude. You can open this up. And now it should ask you to log into Claude and subscribe to Claude to be able to use claw code. Okay, so that's how it works. I recommend just getting a claw subscription. Again, the pro plan is fine to start. Um, but yeah, you can do that. All right, now that we have this here, we have the interface of claw code, which is where we can talk to clock code saying hello. And now I should reply with hello. So the way that clock works is that we're going to be building what I call the agentic workflow. So the reason why it's

[2:53:06] different from all the other builds that we've done is that here we're not going to drag and drop. We're not going to add prompts. We're not going to do all these things. I'm just going to have a conversation, right? And the cloud is going to do everything else for me. It's going to deploy for me. It's going to build it for me. It's going to tap into the software for me. All I have to do is give it all the access and give it the the actual rules and things that I want

[2:53:25] it to do. Um, and then it does it. Okay. So, this is the ultimate level of building agents. It's you having a conversation with AI to build the agent. Okay. And it uses code to like build all that stuff. Now the first thing we have to do is we have to go and open a folder because cloud code in making the agenda workflow is going to make different files. So files is like you know different python files, different code files that are all things that it needs

[2:53:48] to make to be able to make the agent. And so the first step is going here open folder. You can I'm going to add one to desktop new folder and we can name this sales co-pilot and press create and press open. Yes, I trust the authors. Press this again. X and you're good to go. So now every single thing that we do uh or that we make is going to be hosted on our folder. It's very very good because when we exit this we can keep opening different folders. So if I go here to command shift N I can go here I can this and I can open a new folder. Right? And

[2:54:25] so you can have different folders different cloud instances to be able to actually use cloud in different in different ways. Right? So each cloud project is a folder on your MacBook. All right. Right. So now that we installed claw code, the first step is making the form on type form. All right. So I'm here on Typeform. You can just make an account and we can go here to create a form. We can start from scratch or use AI. The AI feature is pretty good. So you can use that. But um multi-question page. Make sure we choose this. And then

[2:54:50] here we can do full name. Actually no. Please fill out your details down below. There you go. Then we can do full name. We can duplicate. Let me delete this. Then this will be company name actually company website and then turn this into a URL down below. We can duplicate [music] this and then we can do LinkedIn

[2:55:15] URL. Duplicate this and then do phone [snorts] number. Phone number. [music] Well, the phone number should be at the top. So we can put it up here and email [music] as well. I believe there is email. There we go. Perfect. That's it. So, we can do an ending page, end screen saying thank you very much. One of our sales reps, one of our consultants, right, will be reaching out to you

[2:55:48] shortly. We can delete this. Delete this. Share nothing. Now, we're good to go. Um, update my timeline. So, new lead inquiry again. So, this right here can be a website inquiry, it can be a Facebook ads, it can be whatever just as long as it's a way for someone to to put their details. And our sales rep, someone who just calls, right, will call the lead with information on the lead and then be able to to be much more prepared for the call itself, which increases the close rate because they have a lot of information uh pre pre- call. Now, let me make this all required

[2:56:25] that they have no way of not filling these out. There we go. Required. Share. And now we're good to go. So, now we can go to clock code and start prompting the agent. Um, I believe so. So, we made the form. Let's press share. Great. So, if I go here, I can put the form. I can see that it's very very easy form. That's it. This. All right. Cool. Now, we have to make the Google sheet. So sheets new

[2:56:52] and this can be full name. This can be phone actually email, phone number, company or website, LinkedIn URL. All right. So now here in terms of sales reps, right? When we research the company, they would want to know the niche or industry of the company based on the website. The primary service based on the website, the ICP fit score. So that's going to be 1 to 10 based on whether the person actually fits our ICP. And then ICP fit reason. So why did the score, you know, go like that? And

[2:57:33] then we can do personalization. Personalized ice breaker. Now, because I'm a bit weird with this, we can turn this into white. And there we go. All right. Awesome. So, now that we have this here, this is going to be the place where the actual sales rep will look at the information before calling the lead, right? And we'll have here, right? So, we have the form and we have the Google sheet. There's one thing that's missing, which is the fact that when it needs to scrape the LinkedIn URL and it needs to scrape the website, and by the way, scrape means that it has to go inside,

[2:58:01] take all the information, and then use it. It has to use some sort of software to be able to do that. So, the best software that I know up to date is a software called apifi.com. So, api.com is a it's like the Amazon for scrapers. I've made tons of videos of Apify and it always keeps updating. As you can see, we have a new MCP connectors um that we can use. Think of them as products. So, Amazon has products. Apiy has scrapers. We have Instagram scraper. We have Facebook scraper, Google Maps, Tik Tok website, which is the one we're going to use [music] um and also LinkedIn as well. So, make

[2:58:30] an account on Apify. And once you make an account, you will get $5 of free credits. All right. All right, so we just logged in here. Uh, we can see that we have $5 of free credits. I already used 43 cents cuz I was testing before before here. U and we also have memory. Now, how does this work with pricing models? Each scraper will cost us something, right? And I'll show you what it will cost us for these scrapers that we're going to use. Um, so we can go to website scraper and we can use this right here because API made it. So, it's very reliable. And

[2:59:02] so the way that this works is that we basically have to give it a URL of a website and then it's going to scrape all the information which is great. So on the top we have the name, we have the the [music] unique kind of ID. So the unique um kind of identificator of clock code you know actually using API and then we have the description. So we have crawl websites and extract text content and for feed AI models LLM applications back to databases or rag pipelines. The actor supports rich formatting using markdown, clean HTML, download files, integrates well with lang chain, llama index, and the wider LLM ecosystem. Now,

[2:59:34] that made pretty much no sense to me um in terms of all these names. I cuz I'm not using them, but we just need to give the website of the person, take all the information, give it to AI, and then AI makes the brief, right? So, we can just copy this, copy this right here because we're going to give AI that exact a scraper. And now that is for the

[2:59:54] website. That is the first thing. [music] The second thing is going to be for the LinkedIn. So, we can go to LinkedIn profile. There you go. We can look at the first one here. We can just open two of them. Let's see which one's the best. Or three or four actually to see what the difference is. We have

[3:00:11] LinkedIn profile scraper plus email. Okay. The second one is LinkedIn profile search. No, I don't need this because we need to give it a URL of LinkedIn profile. LinkedIn post. No, I don't need the post. LinkedIn profile search by name scraper. Nope. Perfect. This right here is what we're going to use. This is good for us because we need a scraper, some sort of system that allows us to give it a URL. So, of my LinkedIn. This right here is my LinkedIn. And basically scrapes this. It scrapes this and everything else as well, right? Like where I worked, where I went to school, and testimonials and all that stuff,

[3:00:45] right? Uh which is great. So, that's what we're going to use uh to do that. Uh, as you can see here, profile details, no email, $4 per 10,000 results. So, for 1,000 profiles, you spend $4 and you have $5 of free credits. So, it's more than enough for most people. So, you can copy this and paste it somewhere else. And now we can talk to claw code. All right. So, now that we're done with API and Type form and Google Sheets, we set up all the foundations for the system because I knew what I wanted before. And typically, that's always what I do. I always kind of make the assets and then

[3:01:14] I tell Claude or the system itself what I wanted to build because I don't like to do things midbuilt, right? That's it. So let's do it. Hey Claude, um you are an expert agent workflow expert and you help people to make agent workflows. Now here is the agentic workflow that I want you to build. So we have a type form which is a software that we use to make the form and whenever the lead fills out the type form I want you to take the LinkedIn URL company website um and then feed it to Apify which is a software that we use to scrape and I'll tell you

[3:01:46] the two different scrapers that we want to use for website and LinkedIn and I want you to extract all the information that we can from each one. Now the form itself that the lead is going to fill out will include full name, email, phone number, website, LinkedIn URL. What I want you to extract on the LinkedIn profile and the website is the niche and industry that they're in, the primary service that they have and whether and also ICP fit score 1 to 10. An ICP fit reason meaning why you score them one out of 10 based on what we do and how

[3:02:14] well they they're able to work with us. So whether they are a good customer for us and then we have a personalized icebreaker that our sales rep can then say um to to the lead. Right? So again, this right here is a sales copilot agent. And this sales co-pilot agent then needs to send all the data to a Google sheet that we have as well. So let me know if you have any questions first before making anything. Then we will make a plan. Then we will create the cloud MD file and then we will uh build the agent and then we will deploy it step by step. Okay, first of all, any

[3:02:45] questions before we start and we can go from there. All right. So you saw exactly um how we can prompt Hey God. Jesus Christ. Hey Claude can change different things cuz I'm using Whisper Flow and sometimes Whisper Flow is not the most accurate which is fine. So I basically give it the description of what I wanted to do and then I told Clawo to make a cloud. MD file once we have things set up. Now what is a cloudd file? Well, I think this explains it really well. Tired of repeating yourself? Tell Claude to remember what you told it using claw.md. It will be a file that is sort of like the main

[3:03:19] system message or kind of like context behind what we're building. So every single time that we run a new session that we have a new conversation with cloud code, it will use a cloud MD file. It will read that get context and then it will answer our question. So much better than just like you know randomly um answering your questions. Now I'm going to turn this from edit automatically to plan mode. These are the different permission modes of clock code. Effort is high. You can max is a bit mad. I mean, the higher you get, the more tokens you use. I think high is fine. I'm going to use plan mode just

[3:03:49] because I wanted to plan first. Bypass permissions is where you want claw code to just do things without, you know, having to ask you any questions. Edit automatically is the default one. And then ask before edits is when it asks you a question every single time that it wants to edit or do something, right? In this case, plan mode is the first one

[3:04:05] we're going to use uh to plan things. And then it's going to start building different things. Okay? So, let's press run. And now it will start marinating actualizing. And we can also see the amount of tokens that we're using which is not the best when it's increasing like that. All right. So now it's going to ask us different questions. [music] So platform question your offer ICP

[3:04:23] question API and keys in Google sheet. How do you want this agent to be built and run? This is the biggest architecture decisions and affects everything else. Do I want this to be an end workflow cloud agent SDK app plain script plus Chrome web hook? I'm going to use this right here. So I wanted to use the APIs of Google Sheets, cloud, API and type form directly with no platform but I maintain it right. We're going to maintain it and deploy it as

[3:04:46] well. Super super easy. It uses Python. I don't know how to code but we have cloud code that knows how to. So so you'll be fine as well. [music] Your offer on ICP to score ICP fit write the ICP reason and craft the icebreaker. The agent needs to know your business. Can you provide it? So we are a marketing agencies that helps coaches and consultants to be able to get more leads through content and we help it's a done for you service and it's mainly for CEOs and founders who are posting actively and want to generate more um leads using social media. If it needs more context

[3:05:22] it's going to ask me so it's fine. I can press enter. Do you already have two API scrapers picked out plus your API keys? Um, I have them ready. So, you can put I have them ready, but I'm just going to put other and just add them here. And I'm just going to paste LinkedIn scraper name and website scraper name. And then also for the API

[3:05:40] key, we can go to Apify. We can go down to settings. We can go to API and integration and we can copy this here. Now, if you're new to API, you can just press create a new token and then you can just put create and it will create one. You can just copy. Okay. Bring it back and put API key. Paste it here and press enter. What should the final Google sheet look like and where is it? Uh let me go here. I can copy this link. I can name this lead in sales coilot

[3:06:18] sheet. There we go. copy the link and then I can paste it to others sheet URL and press submit answers. Now with that said, as you can see, if you guys are never really used claw code, it's really really powerful because everything in the back end is sort of like tapping into different APIs, reading your project, thinking for you, and using AI to the best of the ability to be able to summarize and think through what it is that it needs to do, the task, and basically making an action plan before it does anything, right? And so all we have to do our real skill set here when using cloud code is

[3:06:56] our ability to tell Claude what we want. All right. So now it's asking us how should the agent get triggered when a lead submits type form and where uh will the script run. We can do a web hook. Again a web hook is a is a URL in the server that basically activates whenever you know let's say a type form submission comes in. So a lead fills out the form. It sends the information to our agent using a web hook. Right? A lead fills out a Facebook ads form. It automatically sends the data to our agent using a web [music] hook. So web hook is something that we use to send

[3:07:26] data automatically. So I can use real-time web hook which claw model should power the extraction ICB scoring and icebreaker writing. We can do clotsson 4.6 and press submit answers. So as you can see I'm just directly talking to cloud. I'm not really making anything. Um which is awesome especially when building. All right perfect. So now we just made the plan. We can go here and see it. Uh sales cop pilot build plan context. We are building an automated sales call pilot agent for marketing agencies that runs done for you content service helping coaches and consultants [music] mainly CEOs and founders. When a lead submits a tie

[3:07:56] form, the agent must automatically capture the lead's details, name, email, phone number, website, LinkedIn, and URL. Um, scrape the LinkedIn profile and company website using API and use Cloud Sonic 4.6 to extract and derive the niche, industry, primary service, ICP score, fit, reason, and personalized icebreaker that the sales rep can then use and append everything as a row in a Google sheet. Then it basically tells us, you know, a bunch more things that we don't really need to uh know now, but it's all around the like what files is it making? Like for example the Google sheet writer right sheets py py stands for python and this is the file that

[3:08:28] it's going to make here and it's basically structuring hey here's x y and z that we have to make and kind of like structure so it has a build phase it has a step-by-step plan before it does anything so I can say now yes auto accept I can change this now to bypass permissions so now it's going to start tracking and start building so as you can see now it just made different phases for the product which is very very very good uh because we can then see exactly what step it's at. Phase

[3:08:54] one, scaffold, creating project files. Phase two is writing the claw.md file. Phase three is doing this. Phase four is this. Phase five is this. And phase six is this. Right? So, it's going to go through everything step by step for us to also see what it's doing when it's doing it. All right. Perfect. So, after a couple of minutes, it finished A through F. It made all the files on the left hand side as you can see. Um, you don't necessarily have to know exactly what these are, but they're essentially different files that, you know, are able to do different things. So, Google Sheets is for Google Sheet. API is to

[3:09:22] scrape. APify, analyze is to analyze the actual data that it gets. Um, which is great. And the reason you don't have to know this is because you actually don't need to know what these files are. You can always just ask claude and ask it, hey, what what did this do? Uh, but it's not necessary. So, and then obviously we

[3:09:36] have the main claude.md file, right? This right here is the main kind of the main the main thing. Uh, that goes through sales copilot. So you are this who we are as a company ICP pit fit scoring um scoring brands and then we have ICP fit reason and then personalized icebreaker and then output contract right all right now uh we give it everything I believe now it's asking us to rotate the API key we don't have to do that so it's fine we have to add our anthropic API key for it to use claude to then analyze Google service account create one download JSON next to

[3:10:06] project and share we have to make a Google service account confirm type from fields Okay, so first step is anthropic API key and we have to add it to our EMV file which is this which contains all our API tokens, passwords and all that jazz. So we have to get two actually we have to get the type form and we have to get the anthropic API key. So to get the anthropic API key you can go to platform.claw.com/dashboard you can go to API keys. Uh make sure first of all that you are I believe it's credits that you have some money here because credits do require money but

[3:10:40] it's very very little. So you can have like 10 bucks and it can last you a month depending on how much you use it. So create an API key. We can name this sales co-pilot. We can add and then we can copy this. We can bring it back. Paste it here in the MV or you can simply just put it in the chat. Don't do that because if you put it in the chat now it's basically out there. Um but if you put it directly into the file then it's faster and it's not faster but it's more secure. So you

[3:11:07] can go here you can paste the API key. Now we have to get the type form web hook secret. So I don't know why we need that but let's not do it for now cuz it says optional, right? If cloud really needs this then it's going to ask us. So it's fine. Press command S. So it saves it. Uh now we have to make our Google

[3:11:25] service account. How do I make a Google service account? Open consular cloud. Pick a project. Enable Google Sheets API service accounts. Download the JP. Okay, cool. All right, let's do it. So, let me copy this here. Let me paste it into the browser. All right, so you get here on this page. Make sure you're logged in with your actual account that you're going to use for the Google sheet. Um, you can make a new project. Just name it Copilot. I personally cannot do that just because I have too many projects here and I've reached my limit. Um, but let's say I'm going to use the lead agent. Okay. After

[3:12:00] you want to go to API and services, which you can also find here, API, uh, no API services. There we go. You want to look for Google Sheets because this way we're basically authorizing and saying, "Hey Claude, you have access to my Google Sheets." So it's fine. Google Sheets API. There we go. Right here. And you want to enable this. So there should be a button called enable. Once you press this to confirm, press yes. Wait a couple seconds and then you will be [music] able to see manage. Once you see manage, it means that you know you've enabled the API. Then it tells us to go to the

[3:12:30] service accounts. Uh I am admin service account. Create service account. Okay. Go here or here. I am an admin. Service accounts. We can look for create a service account. And then we can name this sales co-pilot. Create and continue. Is there anything that we have to do specifically or she writer sorry change this to she writer the name. Let me remake this. She writer

[3:12:56] and continue. Continue. Done. Okay. Where is it? [music] She writer. I can press in this. And once I press on this, we have to then go to key, add a new key, choose JSON, and press create. And then we have to get the secret key, which is a JSON file. Okay, so this if this is your first time doing it, don't worry. Just it's just follow this step by step because it's kind of what we have to do. I don't really fully understand it, but I'm I just know that I have to give access to claw code with my Google sheet and I'm just following the instructions. I think it's good that

[3:13:32] I don't really understand it because most of you guys don't, which is perfect because then you can, you know, kind of follow along and see that you don't really need to understand it to do it. Um, create a key for she writer. Again, go to keys, JSON, press create, and I believe now it will make the file. There we go. You have a JSON file here. This JSON file is going to be the one that we're going to use. I believe we have to paste it onto here, but it's fine. What I'm going to do is I'm going to add the file of JSON inside the folder that we

[3:13:59] just made. So, I'm looking at my desktop and I'm going to put this all the way inside the file. [music] And now it should be somewhere here. There we go. Carbon Stone. And here we have the ID, key ID, authorization, all that stuff. Um, again, all I did is I dragged the file into the actual folder that I made this in. So, you can look at Sales Copilot in your desktop or wherever you made the folder. Just drag and drop it there and you'll see it here. Um, and that's it. So [music] I can say now okay so I added the anthropic API key in the

[3:14:33] MV file. I made the Google service account. I added a JSON file called carbon stone 498420J3E7729B whatever. Uh you can look at that JSON file and then use whatever you need for the Google service account. [music] And for the type form fields we're asking the full name, we're asking the email, we're asking the phone number, we're asking the LinkedIn URL, sorry, the

[3:14:55] website and then the LinkedIn URL. That's it. Okay, cool. I press go. Okay, cool. So, now it's asking me to actually add this to the Google sheet and put editor. I can go here to the Google sheet. I can press share. I can paste this and then [music] you can press send. Bear in mind that now you're asking where did I get this? I got it from Claude. But in the JSON file that we just made, if I go here, it's going to be uh I believe this one here, client email. So everything here to here, you can copy this and then put editor um on

[3:15:27] your Google sheet, which means that now it's able to access and edit our, you know, Google sheet file. It's done. Now it's going to test it. So let's see. All right, perfect. It says connected. There's already one row in the sheet. Let me check what it is. Then run the full pipeline end to end with a real profile. Okay, so if you guys remember when I mentioned something around API, right now API is is a thing that allows

[3:15:47] a software to speak to another software. In this case, claw code what it does in you know behind claw code is that it's now tapping into the APIs of these softwares and doing things for us so we don't have to do it. I just say, "Hey, connect Google sheet to Ty form." And then it just goes to Ty form. It goes to Google sheet through the API. It connects both of them and we're good to go. Right? And I'm just here kind of controlling and overseeing um as if you would be the CEO of a company overseeing your employees, right? You wouldn't do the work, but you would look at them, do

[3:16:12] the work, and then you would tell them, "Hey, this is wrong, this is right, great job or whatever." All right, so now I just said it's working end to end. He just made the actual Bill Gates uh example website LinkedIn URL of Bill Gates ICP score 3. Bill Gates nominated founder and decision maker. He's one of the most active content creators on LinkedIn. Okay, perfect. It also made a personalized icebreaker. So, your [music] recent LinkedIn article laying out exactly what the Gates Foundation aims to accomplish over the next 20 years was unusually direct um for philanthropic communication. Most organizations stay vague, so the concrete framing stood out. How closely

[3:16:46] are you personally involved in shaping the editorial voice behind Gates notes? Yeah, definitely not. That's not going to happen. A sales rep would never say that. So, we have to change that. Before doing the last step, which is going live, can we change the personalized icebreaker? It's too formal. Um, I wanted to make it more direct, no fluff, make it in a young tone, as if a young guy is saying it, you know? Um, so yeah,

[3:17:10] change the tone of voice from that. Okay, so now it's telling us to do the last step though. While it's taking care of this, [music] we have to expose the service so Ty from can reach it. Terminal one start the server terminal two in Ty from your form connect web hook ad. I'm not going to do anything. I'm going to ask it what it means to tell me everything step by step. All right. So now I just made the tone of voice different. You post daily on upbound and high ticket closing. Is that content actually driving inbound for you right now or is there a brand play where

[3:17:33] you work leads through other [music] channels? That's very very good. So um well much better than before. All right. For the last step, go live and connect to type form. I don't want to do any of the terminal stuff too complex. Um just give me tell me exactly what I can do in type form to add. Yeah. What I can what I can do there in that platform to keep it as easy as possible without me having to use the terminal. So sometimes cloud code does give you like the more complex solution to do things. Uh which not the best but you know we kind of have to

[3:18:00] tell it hey relax and give us the easy solution so we can just do it. All right. So the agent needs to live online 24/7 so tap from can reach it. How do you want to host it? Render railway. Uh we can do render. I've never used render. So, we can do that. I'll kind of have to figure it out. Submit. Okay. So, render, we have to use GitHub. So, GitHub is a place sort of like Google Drive where we store all these different files. Um, so it's cloud to GitHub, GitHub to render, and then we use Mod as well. So many different softwares. Uh, Mod as well to to actually deploy the

[3:18:32] thing and and make it work. So, what we're going to do now is we're going to go to github.com. We're going to make an account. And then [music] you can press new repository. So basically a new Google drive folder. Name this sales copilot. Press create repo. And then we have to use a link. Give clock the link. Here's the GitHub

[3:18:58] repo link. And we can just press run. So in the GitHub repo, it just pushed it here. So it has all the different files. Perfect. Now it's telling us that we have to go to render.com make an account. So let's go to render.com. I've never used this software. So well I have done actually once. So you can get started. We can go to Google. Continue. And then I believe

[3:19:22] it says new top right blueprint. New blueprint. Connect to GitHub. Choose a repo. In this case sales core pallet connect blueprint name. Does it tell me what the name is? I'm going to make a name here. Sales co c co-pilot. And then we have to add these API keys. So I can go here.

[3:19:49] I can copy this. Paste it here. API token. Paste it here. Google Sheets ID. Paste it here. Tie from web hook secret. I don't have it so it's fine. Uh, and I can press deploy blueprint. And now it's creating the web service. And we should have it up and running. All right. Perfect. So, it just created the web service. After a few minutes, I can go back here. And what's the next step? Um, add your Google key as a secret file. The Google

[3:20:19] login cannot be a normal box in render. Open your source code service. Left menu. All right. So, we can go back to the dashboard. We can look at the projects. We can press here. We can go down to environment. And then we'll see here. Perfect. Secret files. Um, and now I have to add the file here. I believe secret files add secret file. Okay. Plus then the file name if we can put this and then the content. I'm going to paste whatever was in the JSON. So I can find it. I can go here. I can copy everything and I can paste it here and press done

[3:20:54] and press save, rebuild, and deploy. Once you're finished with this, get your live URL. At the top of the render service page, you'll see a URL that looks like this. Test it. Open the URL with health on. Okay, perfect. Okay, so we have to wait for it to redeploy, I believe. Uh, because that is just how it works. So, blueprint source code pilot pilot is a source code

[3:21:17] pilot xxxon render.com. So, I believe it's this one. Oh, there we go. So, pilot zeno. So, we're going to wait because it's still deploying here. All right. All right. So now we can copy this and we have to add a web hook URL in Typeform. [music] So I can go to Typeform. I can go to connect. I can go to web hook here. Add a web hook and I can paste this. And then we have to add the

[3:21:42] /wehook/typeformm. Save web hook. Turn this on. And then [music] it's good to go. Click submit test. Okay. So we have to go here. Redies send test request 200. Perfect. Great. I'll also just paste the URL in clock code and it's saying live and healthy. All good. And in just a few moments, we should be able to test the whole thing end to end. So, when it comes to building agents and stuff, really the the longest part is not so much the building, but it's more everything around it, which is the structuring, the prompts, the you having to communicate what you want, you going back and forth,

[3:22:19] you giving it feedback, you deploying it. But the actual build is just a few prompts here and there connecting it to different softwares and it's pretty it's pretty straightforward after that, right? Uh which is great. All right, perfect. So it says the sales co-pilot is fully live. So now we can test it end to end. So I'm going to go here to type form. I'm going to go and get my URL from the content. Paste it here. And

[3:22:41] then I'm going to get my LinkedIn URL. And I can use my uh thing as an example. LinkedIn URL. Um 2015 555 1 2 3 4 company website. Let's just use acquisition.com website. See if that works. It's going to be inaccurate, right? Cuz my cuz his business because Alex Mo's business and what we do is like completely different. So um Tori

[3:23:11] from my email here and press submit. Thank you very much. one of our consultants will be reaching out to you shortly. All right, perfect. So, we can see here in the logs that it actually ran. It went through the whole process. And in Google Sheets, we can see that we have Mikuel, my email, phone number, website,acquisition.com, a agency coaching, 101 coaching, uh 8 for ICP score. Mikuel is the founder and active coach consultant with running businesses, the a automation circle and jam solutions, an AI agency, uh consulting for agencies, making him the textbook ICB match. He's flagged as a LinkedIn creator with 33,000 followers and he's clearly building a personal

[3:23:46] brand through consistent content, which is a single strongest signal we look for. He's a sold decision maker and has demonstrated revenue traction, 2.5 million in client results. Let's get it, guys. 20K trained um 10K members and community, the one flag, he submitted acquisition.com website instead of his own, which is almost certainly a form error, but means we have no website data to evaluate the business skill dependently. Very, very interesting. So, it understood that this is Alex Ramos's website. I don't know why I didn't think it would do that, but uh that's interesting. And we have the personalized icebreaker. You've got 33K followers and you're clearly posting

[3:24:19] hard about AI agency space. Is LinkedIn actually driving your one-on-one coaching clients right now or is it mostly coming from the school community? As a coach, um like on the coaching side, it is pretty good as in it's a good question to ask. Um and this is a very very interesting piece of information that the sales rep would know before, you know, giving me a call and so on. If you go to Apify here, we can see that the website content crawler was running and it took 41 seconds and the previous one took four minutes and 28. And then we had the LinkedIn profile scraper which also ran for 5 seconds and

[3:24:54] it scrape my profile. Right? So we can see that everything ran successfully. Again, this is a bit different to show the result because it isn't an automation where you can see all the steps kind of running at the same time. It's more of everything in the back end. So we have render which is hosting the actual uh we call the web hook again the thing that triggers the automation. We have tap from here which is going to be the software that actually starts the automation and then we have cloud code which is a place where you know we actually build the actual agent itself and then deploy it and then whenever we

[3:25:22] have to change anything we can go to the cloud code and then say hey change XYZ thing about the system and then change it on GitHub and then from GitHub we change it on render. So, it's a whole process. A bit more nuanced than than the other ones that we've done so far, but super super good use case for any kind of sales team that is able to um get information about their prospects

[3:25:42] before they make a call. All right, so you've now built five different agents on four different agenda platforms and each one of those is something that a business would actually pay for. Now, you can now build things, but building is the easy part. The hard part is taking this information and this technical knowledge and turning it into a real business and [music] essentially getting someone to pay you, which is exactly what the next segment is all about, which is segment [music] three, monetizing. The first thing we're going to look at is picking one opportunity and one offer. All right, so the biggest mistake that people make

[3:26:16] getting into the AI space and wanting to build a business is that they focused on the wrong thing. which means that they get into the AI space and they start learning the tools you know watching videos on clock code and it's an Hermes agent whatever it is and from there they start thinking okay what can I build with those tools and then they say okay well who can I build it for and then what is the offer and then what is the problem that I'm solving for the person and the reason that's the wrong way to approach things is because business all it is is taking a customer from point A

[3:26:43] which is where they are right now missed leads you know less calls hello conversions all the way to point B which is where they want to Fewer mis calls, higher conversions, more money. And so you as an individual, as the operator, you need to get the customer from that point A to point B. And the car is the tools. The car is the vehicle. But the vehicle doesn't matter if you don't understand what point A point B is. And so most beginners spend months, if not years perfecting the actual vehicle when they should be perfecting and understanding the problem that they're solving. So with that said, let's dive

[3:27:13] into the different problems that all the five systems solve. Starting with website lead generation agent. So this solves the problem that website visitors leave without converting because there's nothing there to engage, qualify and capture them in the moment and each lead is potential revenue. The second system is the inbound receptionist. So businesses miss calls and the revenue attached to them because there's no one available to answer, qualify and book every caller around the clock. The third

[3:27:38] [music] system is speed to lead. Businesses lose ad leads because nobody calls them fast enough. So the lead goes cold before they're ever qualified or booked. The fourth system is executive assistant. So owners and operators drown in admin whether it's email scheduling contacts instead of spending time on the work that actually grows the business. And last system [music] is lead research sales copilot. So sales reps waste hours manually researching prospects before calling instead of actually selling. And so with that said, those are the exact main problems that we're solving with each specific system that we built because you knowing how to build a system, that's the first step and it's

[3:28:11] only going to get easier to build. But you understanding the problem that it solves is the most important part because then [music] you can think of who has the problem and you can go and sell this. Now my point here is that you have to pick one. So pick one specific system that we built and we're going to build an offer around it. So for example, let's say that we choose a speed to lead. We're not going to sell speed to lead. We're going to sell the

[3:28:30] outcome that a speed tole provides. [music] So, instead of saying, "I build speed tolead agents," we're going to say, "I help you build speed to lead voice systems that qualify your ad leads in under five [music] minutes and book them into your calendar without hiring extra staff or changing your ad setup." And the reason it works is because it's very outcome driven. Do I say AI in the offer? No. Because businesses don't care about AI. They don't really know what AI even is. And so, you speaking in words of AI makes them confused. [music] And a confused buyer never buys. Hey guys, quick one here. If you're enjoying this

[3:29:00] video and you are someone who actually wants to take these skills and turn them into a real AI agency by working with me onetoone, then check out the first one down below which walks through a full video on how everything works. Now, let's get back to the video. All right, so now that you've picked one specific system and you've built an offer around it, again, speaking outcomes, not tech, you now know what to sell. But the next

[3:29:20] question is, who are we selling this to? Because the same agent is not worth the same for each business. And so that is the market. So when it comes to talking about niche, it will make much more sense once we break down the pricing structure of these systems. So fundamentally, one of the biggest things that I see in the space is that people talk about pricing based on the hours

[3:29:39] that it takes you to build the thing. I'm very against that just because of the fact that you will get better the more you build and it will take you less time the more you build. So should you get paid less because you're better at the thing? Of course not. Right? And so we want to charge based on the value that we create for the customer, not the time that it takes you to build the thing. And this is the trap that I fell into very early on because I remember pricing a system 500 bucks because it took me 3 hours and I came up with a

[3:30:06] hourly rate out of nowhere and end up making the business over $8,000. And so if I had just charged based on the value, I could have charged two to three times more than what I did. But it was because I was obsessed with how long does it take me to build the thing? Is it easy? is a complex that I just charged based on how long it took me to build the thing, not to how much money you made the business or save the business. So with that said, do not make the same mistake that I did and charge based on the value that you create. So

[3:30:32] let's use an inbound receptionist as an example for this. An inbound receptionist solves the problem of missed calls after hours. To understand the value that we're providing for the business, we have to understand how much money we're making the business or saving the business. In this case, let's say the business tells us that they're missing 20 calls a week. So that will be 80 calls per month. And let's say that they also tell us that each customer for them is worth just around $2,000. And so you would do 80 * $2,000 which is around $160k. And so what you would do is you would charge anywhere from 10 to 20% of

[3:31:05] that which is 16k to 32k. Now I know that sounds extreme to you because maybe you've never sold anything that high ticket. At the same time though, when you break down the numbers in front of the business owner, it will make sense for them because they can see the extreme ROI that they get from your system and [music] the price, the investment is nothing compared to what they get. But if you had just priced based on how many hours it took you to build the thing, maybe you would have charged 2,000. And so instead, we're charging 8x more because you're focusing on the right metric. So to conclude

[3:31:33] this, the more money we can make the business, the more money we can make. So when it comes to who we target, we have to understand which niches can I provide the most value to? Can I make them the most money so that I can make the most money? And so a lot of beginners when it comes to niche, they pick a niche

[3:31:49] randomly based on what they think works. And then they wonder why they can sell this agent for 100 bucks versus 2,000 bucks. And so here we're not going to pick a niche randomly. We're going to pick the niches where a single lead, a single missed call is super expensive for the business. Because a missed call for a restaurant is only going to be worth 50 bucks, but a missed call for a dental clinic, a med spa, a roofer, or even a solar company is hundreds, if not tens of thousands. And so we can just charge 5 to 10x more just by changing who we work with. And so the people that

[3:32:20] you would target will be med spas, dental clinics, home service, marketing agencies because the value of one lead is high enough that solving the problem is worth real money for you. And this framework right here of pricing is the one that we teach to all of our students that work with me onetoone because it is by far the most effective and allows them to charge anywhere from 5, 10, even 20 and 30k as well instead of charging like most people which is 500 bucks, 1,000 bucks or just 1.5. All right. So now that you understand who it is that we have to sell to and how to price, the

[3:32:52] next question is how do I get my first few customers which is where we get to the next section which is warreach and free pilots. Now when it comes to pitching the service a lot of people feel like an impostor because they have barely delivered the product in the first place. And so they tell themselves that they need portfolio of businesses, testimonials, a ton of like experience and building and building building. So they watch more videos and they never actually reach out to anyone. But

[3:33:16] realize this, everyone starts from zero. I started with zero clients. And so to get case studies, you have to work with people to then get case studies and testimonials and experience and referrals, right? And so the first part is reaching out to people that you already know. So here's exactly step by step how to get your first client through warm out reach. The first step is going through your phone. Second step is go to contacts. Then look at the amount of contacts you have. And then I want you to divide the actual list into three different categories. business owners that you know, closed group friends and family and other people. For

[3:33:47] the business owners, we're going to call them. For the closed group, we're going to call them. And for other people, we're going to text them, and we're going to tell them all the same exact thing. And by the way, let's say we're using the inbound receptionist as an example. Hey name, I started a business helping dental clinics, med spas, and home service businesses to stop losing customers from miss calls. I'm opening up five pilot spots while I finish refining my system and would also like more testimonials to show it works across different scenarios. Do you happen to know anyone who might be

[3:34:13] interested? Totally cool of thought. Thought I'd ask. Let me know. As you saw from that message, you're not pitching. You're asking a question. Now, here's why this is so effective. Because this right here is you. And around you is your first degree connections, people that you can contact directly. But you might think that you know only them when really there's a bigger circle around the first degree connections, which is second degree connections, which is people that they know. And so what you're going to do is you're going to reach out to the first degree connections to ask them if they know anyone. And if they resonate with your

[3:34:46] message, they will say, "Yeah, you can help me." And if they don't resonate with your message, but they know someone, they can refer you to someone. And so it's a win-win. And by the way, we're not doing this completely for free because we're doing this in exchange for a referral, testimonials, and experience. And so those are much more valuable than getting a 1 to 2k client early on. Now, once you finished the actual warm outreach and you've gone through your whole list, you now have a few clients you're working with. and [music] the referrals as well. You now want to tap into cold outreach, which is

[3:35:12] reaching out to strangers who have no idea who you are. And that's looking at cold calling, uh, cold emailing, LinkedIn DMs, or even networking events. The most important thing is that you choose one, you double down, and you go full in. All right, so now you know exactly how to get these conversations [music] in. But now, how do you turn these conversations into paid clients and actually make money? Which is where we get to the next section, [music] which is sales process, turning interest into a paid client. So, a ton of beginners, including myself, by the way, which is why I'm telling you this directly, when they go on a sales call,

[3:35:43] they overexlain the tech. They say the price when the prospect goes silent, and they talk themselves out of the sale. And the reason that is is because they have the idea of the call, the structure, and the foundations of the call completely wrong. So, to make it super simple, let's break down exactly how you should approach the call and what things you should say. So, the

[3:36:01] initial call is called a discovery call. And this is a diagnosis if call. Meaning that you diagnose their business, their problems, their data, and then say, "Okay, based on this, can I actually help?" If you can't, you tell them you can't. If you can, then you close. And so this call right here is you working out the what outcome they actually need, how much the problem is costing them, and whether you can fix it. Because the price, like we mentioned, is value based

[3:36:26] pricing. Value is attached to money. Money is data, data is numbers, and numbers is questions. So it really comes down to what questions you ask to get the right outcomes to then get the right price. And so here's the shape of the call at a high level. The first thing you want to do is you want to understand their situation, their business, their ICP, and the revenue. Then you want to find the real problem and what it cost them. This is asking questions like how many calls do you get, how many missed calls, what's their current conversion, how much does a client usually pay when they go into your business. Then you

[3:36:57] show you can solve it and then talk scope and money. All right. So now that you've closed the client, the next question is how do you deliver the system for the client in a way that gets them results, [music] gets you referrals, and helps you to retain the client longer? Because a ton of people think that closing the client is the last step [music] when really it's just the first step because you want to make sure that the client has a good experience. You want to make sure that you can then retain them and upsell them, make more per customer rather than having to find 10 new customers every

[3:37:26] single month. [music] So the next section is delivery and retain. So, as weird as it sounds, and I don't think you're going to hear this much on the internet, the client experience matters almost just as much, if not more, than the outcome and the result that they get. Meaning that, let's take three [music] different scenarios. The client has a great experience, but they don't really get a good outcome. The client has a bad experience and they get a good outcome. The client has a bad experience and a bad outcome. And a client has a

[3:37:55] good experience and good outcome. Right? That is a matrix that we use. Of course, a good outcome and good experience, that's green. Bad outcome and bad experience, that's red. But this is the part where things get interesting. You have a higher likelihood of working with the client long term if they have a better experience. Meaning that like of course you want to make sure that you do the best job for the client. The outcome is the best. That is important. But what's just as important for the client is how they feel throughout the whole process. And so your onboarding delivery process needs to be optimized in a way

[3:38:24] that makes the client feel like they are taken care of. So really the experience determines whether they want to work with you long term and whether they give you referrals. So the first step of the delivery process is what we call the onboarding call. The onboarding call is there to set expectations around time and availability and collect all the softers and access that you need to deliver. The next step is building. So you want to send weekly updates every single week and the updates can be as easy as sending an email saying, "Hey John, hope everything is good. Quick update on XYZ project. We just finished

[3:38:54] milestone one. Now we're starting on milestone 2. looking forward to showing you how it works. Talk soon. That way the client stays informed and they don't have to ask you where the product is at. Then we have step three, which is a testing period. So take [music] one to two weeks to test things yourself with your team or with the client as well. If you're building a voice system, you always want to make sure that you get the client input so that they give you feedback and then you can improve the system. And then finally, we have the delivery call, which is where you show them how the system works. And this is

[3:39:22] the call where you're going to upsell them. And so if they had a good experience and you've done a good job, there's almost a guarantee that they're willing to work with you long term, assuming that they have another problem. And so on that call, you want to upsell them, which means that you're offering a client a way to solve a different problem. So let's say that you're using an inbomb receptionist as a gateway offer, then you can upsell them to a speed to lead or a sales co-pilot. And so once you finish the first project, it's deployed, you would say, "Hey, by the way, I know that you have X, Y, and

[3:39:49] Z problem. We can also solve that too." And then you go into the details as to how you can solve it. And if you've done a good job, they already have trust. You have authority, right? You've proved yourself. And so now you have a much higher likelihood of closing the second client. Because I can tell you this, businesses just want the work done. They want the problem solved. And so they would much rather work with one person long term than having to find new contractors and new agencies every single time. And this is the exact difference between a agency and freelancing. Freelancing is one-time products. Agencies is retainers. and

[3:40:22] maximizing the amount that you make per client. All right, team. So, that right there is the full AI agent master class where I took you through exactly what an AI agent actually is. We built five systems [music] on four different agent platforms. And we also went through in detail as to how you can turn this information and these technical skills into an actual business and [music] get paid for them. And so, if you got anything out of this video, the best next step you can do is not watch another video, is to just start. pick one of these agents and actually build it this week because none of this

[3:40:50] matters if you don't know how to sell this. And by the way, I'm going to leave all the resources and templates in the third link down below. You can go to my school community, you can go to the classroom section [music] and you'll find everything in the templates vault. And if you're someone who is working 9 to5 and you want to take these skills and turn them into a real business by working with me onetoone, then [music] check out the first link down below which walks through a full video on how it works. Check it out. If it sounds like a good fit, then feel free to

[3:41:14] apply. And now that you know exactly what to sell, how do you get your first free client, and how to close them, check out this video on the screen where I show you step by step the road map that you can take to get your first AI paying client in 30 days. With that said, guys, I really hope you guys found

[3:41:29] value from this video.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt