# Building AI Agents that actually work (Full Course)

Channel: Greg Isenberg
Video: https://www.youtube.com/watch?v=eA9Zf2-qYYM
Duration: 59 min
Language: English
Words: 10674
Transcript page: https://viewrankai.com/tools/youtube-transcript/eA9Zf2-qYYM

---

[0:00] I think AI is confusing. There, I said it. I think there's a lot of terms, skills, MCPs, agent harnesses that are difficult concepts to understand. So, I had my friend Remy come on the podcast and explain it in the most simple terms possible. In this free course on how to master AI agents, he breaks down exactly what each piece is, how they connect together, and the simplest ways

[0:27] beginners could start using them today. Enjoy the episode. [music]

[0:39] I begged them to come on. Remy Gaskill's on the pod. You've structured your company where you basically have these folders and dot MD files that run your company. And what I want to do today is I want you to teach people in a beginner-friendly fashion. This is only for beginners. How they could do the same thing. How they can set up their own executive assistant, head of marketing, chief financial officer. Basically, I want you to base to tell us the concepts behind all this. By the end of this episode,

[1:12] Remy, do you think you can do that? 100% Greg. We're going to go through all the concepts that make up an AI agent. And by the end of this video, you will know exactly how you can build up agents to run complete departments of your life and your company within any agent platform you choose, whether it's Claude

[1:30] Codec, Open Claw, Manas, all of them. All right, let's do it. Sweet. So, one of the reasons why I really wanted to make this episode is because I feel like the AI landscape is moving into like stage two, from chat to agents. And most people are getting left behind right now, just using the chat models. And the founders and employees that are utilizing agents are no word of a lie 10 to 20 times more productive in their day. And when you stack that up over days, weeks, years, you're going to just be miles ahead of the competition. So, I really want to make this episode today to help bring

[2:09] everyone up to where the AI landscape is at the moment and to start using agents to manage every department of your business. So, the key thing to understand here is chat models versus agents because the word agent is thrown around lots online. I'm sure you've seen it, Greg, like AI agents this, agents this, use this agent for this, and it's kind of lost a lot of meaning. So, I wanted to give start by giving a really clear definition of what

[2:34] an agent actually is. So, the way I think of it is a chat model is question to answer. But then an agent is goal to result. So, moving from just like uh you asking, AI replies, then you do the work to you giving the agent a task, it planning out the task and then executing and then delivering you a result. Does that make

[2:57] sense? Crystal clear. I mean, the way I think about it is chat is kind of like ping pong, back and forth, back and forth. Yeah. And agent is uh you know, you're giving it it's a goal. I mean, the best way to Yeah, you're giving it a goal and you're hoping that over time it gets better and closer to

[3:14] that goal. Exactly. Yeah, that's exactly it. And I just think that's a nice way to lay it out in your head is chat is question to answer, agent is goal to result. So, when you chat to an agent, you might give it a task like build me a website for XYZ, and then it goes away, it does its work, and it outputs this

[3:34] wonderful website to you. But it's really important to understand what's actually happening in this step here. So, inside this agent step, we have what's called the agent loop. So, you give it your prompt or task, and it goes through these three steps here, which is observe, think, and act. So, let's just say for example, um we're actually going to do this demo after this, but if we gave the agent a simple task like "Build me a minimalist portfolio site

[4:03] for Greg Eisenberg." It's going to start by like you've loaded in that prompt. It's going to check if there's any files in the workspace that it can work with, like maybe you've got some information on Greg Eisenberg. Um and then it's going to think about what to do next. It's going to act, and then it just keeps going through this loop. So, for that actual example of building the portfolio site for Greg Eisenberg, let's just say it was a blank agent. We hadn't given it any context. The first thing is it's received this prompt to build

[4:32] the website. And the first thing it's going to be thinking about is, okay, well, I need to build this website about Greg. Who the hell is Greg Eisenberg? So, it's going to then decide to do some research into Greg Eisenberg. It's going to research everything about Greg, and then feed it back into this observe step. So, then it's going to think to itself, "Okay, so I've got this prompt to build a website. I've now got my research here, so I know exactly who Greg

[4:57] Eisenberg is." And then it's going to start thinking, "What is the next step?" And the next step is probably to write up a plan to build the website. So, it might write up that plan, feed that back in. Now it's got the research, the prompt, the plan, and it will think, "All right, what next? I should probably write the code." It'll write the code, feed it back in, and it just keeps going through this loop as many times as it needs until it can conclude uh that the task is complete. And how it concludes that the task is complete is based on the parameters that you set in your

[5:28] prompt. So, you know, if you're giving it a research task, you might say compile 10 sources, and then create a report as a PowerPoint. And then once it's compiled 10 sources and build the report as a PowerPoint, it can conclude that the task is complete, and then give you the output as the user. The The agent itself

[5:48] is made up of these four components. So, it's the LLM, which is the brain behind it. So, think like, you know, Claude Opus 4.6 or GPT 5.4 or Gemini 3. It's the model. Uh it's got the loop, which means it just keeps going until the task is done. It doesn't stop after one response. So, you're going from ping-pong to like it continuing to go rather than you having

[6:11] to sit there baby-sitting it. Uh it connects in all your tools. And then it connects in all the context. And a platform that facilitates this process and basically facilitates this loop to happen is known as an agent harness. And all of the popular AI agent platforms on the market that you'd be familiar with are just agent harnesses. They're just applications

[6:36] where this loop is facilitated. And I want to actually run this little prompt I prepared earlier. I want to open up CodeX, Claude Code, and Anti-Gravity. And I'm going to show you this loop uh actually happening in action. So, I have a nicely prepared before the episode these three demo folders, which we're going to run in. So, I'm going to open up demo one

[6:59] to work in in Claude Code. And the way these folders work is if you've used If you're familiar with like any of the chat models like Claude and ChatGPT, there's a projects feature, which is where um if I open it up actually, try not to get dizzy with me switching tabs so much. But, you know, if we create a a project here,

[7:19] it contains all your chats in one place. It allows you to upload all your sources here, which is your context. And then you can even add custom instructions, which tells it how to behave within this project, and that's also known as a system prompt, which we're going to dive into how to do this with agents as well later. But, it's a similar concept that you'd be familiar with if you've used projects before. But, instead of the project being here on the cloud, we're actually working within projects that uh

[7:46] local on our computer. So, I've just selected this demo one for now. Then, we're going to run build a minimalist portfolio site for Greg Eisenberg. And then, this little bit here just tells it to actually spin it up like to publish it on the web uh in a preview mode so we can see what

[8:03] it's done. So, I'm going to run that. So, so this is this is um Claude Code? Yes. Yeah, right now we're in Claude Code, and this is just accessing it through the desktop app for Claude. Um so, I'm just going to run that. And then, I'm also going to give the same prompt to Codex here. So, this is the Codex app. And you can see same concept. It says, "Let's build." We can choose a folder on our computer to work

[8:29] in, like demo two. And then, we're going to give that a prompt as well. And we're going to tell it to host it on a different one. And then, also in Antigravity. So, you can see same concept. We're going in, selecting a folder. And then, we will give it the prompt as well. How should people think about security

[8:54] and these different products? I like to think of security as in just like scoping what they have access to. So, by default, Antigravity, Claude Code, and Codex, they're very, very secure because they're built by these massive companies that have a lot on the line um to protect. And I just you know, if you're if you're building out these agents to manage different elements of your business, like the other week I built one um that does manages meta ads. And obviously, that's quite a risky thing to give an

[9:26] agent control over managing ad budgets. So, it's just comes down to like what you feel comfortable giving the agent. And also, you can control what privileges or you can control what um tool permissions it has access to so that if it was compromised for whatever reason, the worst case like isn't that bad. And that means, you know, just giving it like read-only access to certain important platforms and stuff

[9:48] like that. Does that make sense? Yeah, totally. I mean, comparing it to like open claw, which is like Yeah, which I want to touch on at the end as well cuz that's the same thing, just another harness, but it's just like the wild west. Cool. Uh, and one thing like a nice little analogy to think about these harnesses is what we're going to learn today is we're going to learn to drive. So, we're going to learn about how to, you know, steer the car, like how the pedals, the brakes work, the accelerator works, the handbrake. But then once you know how to drive, you can kind of jump in any car

[10:23] whether it's like a old Toyota, a Range Rover, and you inherently sort of know what to do. And that just comes down to understanding all these key concepts that we're going to go through today. And you can think of the agent harnesses like different cars. And some of them will have better features like seat warmers and cruise control, but it's all once you know how to drive, you can pretty much jump in any of them and use

[10:42] them. So, we've just got our thing over here building to building the website for Greg. And it's going through this agent loop right now. So, you can see here it's actually decided that it's going to launch, um, an agent to go and research Greg Eisenberg. And I've connected it up to Perplexity, so it's now using Perplexity to research Greg. So, it's going through its first step of the loop. And I imagine that, uh, Codex

[11:08] has also done something similar here. You can see it's still working, but it's gone. And, um, started to build this out through the loop. I think Claude code does the best job of actually displaying that loop, um, and allowing you to see what it's thought about compared to Antigravity and Codex. But it's all just going through the same sort of loop process

[11:30] that I described earlier. And I think So, when you say that when you say you hooked it up to Perplexity didn't it's not like you asked it to hook it up, right? It just sort of did it. Yeah, because I've I've given Claude code Perplexity as a tool via MCP. Okay. Which we're going to get into um very very shortly all about MCPs, which

[11:54] is just connecting tools up. So we can see that in anti-gravity it's gone you can see this thinking process. It's gone um I'm now examining the current directory to figure out if there's an existing project or if I build one from scratch. It's then going I'm now going to start to build this thing. And then it's built the the website and it's given us a little localhost preview here. So it's created this nice little

[12:19] portfolio site for for you, Greg. What's interesting is like it's super minimalist and I mean it it did its job, right? Like this is I would totally launch something like this. [laughter] It actually looks really nice. Did it Did it scrape your email address correct? That's not my email address and I don't live in South Canada anymore. So but yeah, there's a few copy things, but other than that uh

[12:46] Yeah. It's done a pretty good job. It did. And if we go So that was anti-gravity. If we go into Codex as well, you can see here it's finished doing its website, which is somewhat similar. I think I prefer Gemini's. Yeah, agreed. If we check out Claude as well um it's still going. But you can see this loop, right? It's gone Okay, so first off, who is Greg Eisenberg? It's gone and researched Greg, then fed it back into that observe step and it's gone, all right, what

[13:18] next? Now I need to create the HTML file. So it's written the the code. And then now it's gone, okay, so he wanted it spun up on this local server. So, now I'm going to spin it up on the server. And then the last iteration of the loop is to check that it's actually done and can conclude the task is complete. It's opening it up and screenshotting the website and then reviewing the screenshots to check that the website is complete. And you can see here it's done another pretty good job. This one's very

[13:48] similar to the Gemini one, hey? It's true. Um but yeah, that's just like demoing how that loop is actually working in real time. Yeah. I mean, what comes to mind just by watching this is like how many people on the planet would benefit from a very clean website and like Yeah. how how do you set up these agents so that like you know, maybe it's like a cold email loop, right? Like you're sending cold emails, "Hey, I built you this website, so-and-so business. Do you want it? It's

[14:19] going to cost $250." Yeah. Yeah, that's actually a great idea. Um pre-making websites for companies. And it's like an off-the-shelf thing. It's like, "Hey, I I made you this website. If you want it like if you want to own it, it's $250." You can just do a mass cold email thing. Uh cool. So, I think that's like pretty much illustrated that agent loop example. So, I'm just going to go um

[14:42] back to our trusty board over here. But you can understand that it's just like all of these apps are just different flavors of the same thing. And then what we're going to be working up to today is my workspace looks something like this is I have, you know, a big like a a folder for each company

[14:59] or or client that I'm working in. And then I'll have folders underneath with all my heads of departments. And then um within those heads of departments, I'll have skills and MCPs, which we'll get into. And contacts. And then I've got like an overarching one at the top to just to sort of manage them all. But we're going to be focusing today on building out this executive assistant to take care of just your manual day-to-day tasks and free up at

[15:24] least 1 to 2 hours extra per day. Um cool. So, to build this out like we did uh with our demos, it's running off your local files. So, we're going to create a folder here called executive assistant. And also through building out this assistant, it's going to allow us to clearly explain each of the concepts um

[15:49] of building an agent in real time. And the way I like to think about building agents is onboarding them like a real employee. So, if you took on a real executive assistant, you couldn't expect to just for them to come into the office and you to give them a task without explaining your business first, your clients, what you do, the tools um because they just would not be a very

[16:11] good executive assistant. So, that's the first step that we need to go through uh when we're building out this agent. So, uh I'm actually going to work uh in Co-work at the beginning. So, Co-work is just another agent harness to do the pretty much the same thing as all the others, just that loop connecting in

[16:29] your tools and the context. So, you can see here um this was my little uh previous session where I was building some diagrams. But, we can go um and you can follow along in Cloud Code or Codex or Antigravity or whatever agent harness that you want to work in. But, I just think that Co-work has really nice simple UI for people to just understand really well what's actually

[16:50] going on. So, we're going to open up this executive assistant folder. And you can see here that if we ask it, "Write me a cold email." and send that off. So, peop- So, people are going to ask, "How How did you transcribe?" You did like a voice-to-text. Yeah, so that is um I use one called Monologue, but there's a lot out there on the market. Whisper flows another popular one, and it just allows you to hold a little button on your computer and just yap away, and it will just transcribe it neatly into text. And I find that

[17:26] good. My monologue looks really good. Yeah, I think it's built by the team at Every. Every. io. Um it's a cool product, but uh so what it's asking So it's it's straight away we we it's got no context here. So it's working out of uh that folder here on our computer, but there's nothing in the folder.

[17:46] Mhm. And it has no memory of our previous sessions. Um and it's asking like what like what do you even sell? Um and then we're going to kind of give it like who do you target? What tone do you want? These are all things that our executive assistant should know. Um so I'm just going to stop the

[18:02] response there. And one thing that's really important to know, which might be a bit of a shock moving from chat to agents, is that these agents' memory work a little bit different. So if you're used to using chat models like ChatGPT and Claude, if you open up a fresh session in the one of these chats, you don't give it any context, you don't upload any files, and you just say "Who am I and what do I do?" It's going to know a scary amount about you. And that's because with these chat models,

[18:30] they have memory built in automatically. So every time you sort of say things that are important, the chat model saves it to its memory in the cloud that you can't see and you can't control. And with agents, you have to set up memory and control exactly what you give it. And I think that's actually uh a benefit, not a limitation, because what happens is if you're using ChatGPT and it's got the auto memory, you're having conversations about three different companies, maybe you're asking for relationship advice, and then all of a sudden um when you ask it to write a landing page copy, it's pulling in context from all these

[19:06] other places that you don't really want in there. So with these agents, um you need to actually set up that context and memory. So, as you can see, when we asked it to write a cold email, it just had no idea about anything. So, we need to give it a context file. And the way you do this, right? So, you can see this example here. It doesn't

[19:28] know anything about us. And that's because we haven't populated what's called an agents.md file. And an agents.md file is just like a system prompt. Just like if you've created any custom GPTs before, you have that field for custom instructions. Or in the project, like I just showed before, you have that field for custom instructions. And it just gives it this context that's kind of always there, always on. And you put in there things like its role, context about you,

[19:56] um, your preferences for working. And then what happens is every new session, before it answers your your query or task, it loads in all this context to its brain as part of that observe step in the loop. So, I have pre-prepared pardon me. So, I've pre-prepared a, uh, agents.md file here. So, if we drag this in over here, this, uh, when you're working within

[20:26] Claude Code, it's called a Claude.md. When you're working within Gemini, it's called a Gemini.md. But when you're working in within Codex or Open Claw, it's an agents.md. But it's all the same concept. So, we can drag this into our folder here. And if we open up this file for a little preview, we can see here, I've got in here, um, all about me, what my business does, my working preferences, like the tools that I use and what for, like Notion project management, Stripe, um, we've got, you know, all the information. My ideal customer, it's

[21:02] loaded with context here. And I pre-prepared this, but if you want to make one of those, you can just use Claude chat or co-work whatever, and you can ask it to help you build out this agents.md file, and to just ask you interview-style questions to extract all the context from you, and then build the

[21:20] file. So, if I jump back in, now if I go to a new task, same folder, and we say, write me a cold email, it's going to have all that context. Yeah. That's what we hope. That's what we hope. There we go. And it knows, you can see these files over here. It knows automatically to load in this file if

[21:48] you title it correctly. Yeah. It's basically just like a a reminder file. Yeah, pretty much. It's just like loading it in so it has all this set context before you even start working. And one of the other big shifts to make, which comes with moving from chat to agents, is prompt engineering used to be the big thing. It was like, here's the ultimate prompt for going viral on social media, or use this prompt for this. And now it's all about context engineering. It's about how well can you load up your agent with all the information about your business, so that your prompt can be stupidly simple, like write me a cold email, and

[22:26] you're still going to get an amazing result. Um you can see already here, it's already asking like, um is it a brand or sponsor, potential partner, or consulting client. So, it's already got that context. Um book a call. Um you know, it's it's it's loaded in everything that we've given it from that agents.md file. And then now we've got a a pretty decent

[22:50] cold email there, ready to go. So, that's basically agents.md files for you. And you want to create one of those to onboard your agent with all the context it needs. And if you have lots of context, without getting into too many advanced concepts here, sometimes what I will do is I will create like a um

[23:12] a folder called context. Load that in. And in here it's got different files about me, brand voice, ideal customer profile, etc., etc. And then in order to keep this smaller, I will then just say in this Claude.md file, um before answering any questions or before doing any tasks, read my context folder to understand about myself and my

[23:35] business. Because by default, if you just have this context file in here but no Claude.md, it won't load all that into the session by default. But if you tell it in this file that it always loads in to then check this file, you can start to like string all your context together. And a lot of people have done that with Obsidian. So, they'll have like in the Claude.md file, they'll tell it to go check their Obsidian vault for their

[24:00] second brain to go and find context. Mhm. So, that is agents.md files explained. So, that's how you actually when you're onboarding your agent, like our executive assistant, you can train it up on who you are and your business. And then as you can see here, you know, I've got folders for all these different roles in my business. And in the head of marketing, that Claude.md file would look somewhat similar, but in the top it would say instead like you are my head of marketing. You speak like this. These

[24:27] are your tasks. These are your roles. And then the second thing here is about memory and the self-improving loop. So, we've solved the problem now. Um try not to get too dizzy with me switching tabs. But we've solved the problem of our executive assistant not knowing anything about us or our business. But now we have a new problem, which is it doesn't really remember the intricate details or your preferences across sessions unless you're manually going and updating that

[24:56] Claude.md file. So, you can see here if we go um my favorite color is lavender. It'll probably say something like got it, noted. Yeah, that makes sense, right? Cuz it's and it's it's adding Where Where's that adding it? Well, it's not adding it, that's the thing. So, we can tell it my favorite color's lavender. And it's gone. The user's just shared you That's that thinking step. It's like the user's just shared this No like no nothing needed. Good to know, I'll keep that in mind. But then if we go into a new session, same folder, and we go what is my

[25:38] fav color? Mind my spelling. It's going to say I've no idea what your favorite color is, even though we just told it. And that um is an issue, you know, cuz if you're working, you know, in uh you've got like a head of sales or something and it keeps it signs off your emails wrong. And you tell it you correct it, you say never sign off emails with cheers, say

[26:00] warm regards. And it will go Okay, got it, noted. But then the next day you start working and it does the same thing again. It's like, what like my agent's broken. But really it's not it's running off those context files in the back and unless you are manually updating it, it won't know to

[26:17] save that preference. So, what I like to do is I like to add in something like this to my agents.md file. So, this is just a little simple thing. You can pause the video and copy it. But I like to I'm just going to remove that context file for now. That was just to illustrate that example of adding more, but we're just working with this one file for now. So, I'm just going to open

[26:44] this up so I can edit it. And I will quite often add something on the bottom like that little snippet. And this basically just says Actually, you know what? I might just add it at the top. Just so it's there. Top of mind for my agent, cuz I think this is really important. So, you can see I've just added this in and it just says

[27:07] read all files in context. Read memory.md. This is what you've learned over time. And then when I correct you or you learn something new, update the relevant section in memory.md. And it's just got a couple little things here. And it just says keep memory.md current. When something changes, update it in place and replace outdated info.

[27:29] So, we can do command S to save that. And then I'm going to add another file here. We can actually just duplicate this. And this one I'm going to call memory.md. And then we can open up this one. And I'm just going to remove all of this context here. Um except I'm just going to keep those

[27:53] sections. So, memory.md is basically I mean, it's just what it sounds like, right? It's basically like you want to [sighs] you know, if the goal is to build, you know, AI employees that do things for us, they're going to need to need to remember our preferences, right? A good employee remembers preferences. Exactly. And learns over time and that that compounds. So, memory.md is just a place that you can just make sure that uh over time it you know, your whatever you're using co-work or whatever, it ends up it ends up getting compounded, getting smarter. So, ultimately you might be trying things like co-work and and you're you're not getting good

[28:37] results. And a big part of that is you don't have uh a clo.md and memory.md sort of Yeah. set up. Exactly. Exactly. And now the thing is some of these agent harnesses have started to add in this memory system that we're doing manually, telling it to update. Some of them have have got that built in automatically, like Open Claw, and I believe like Manas and some of the others have that built in automatically. But it's still important to understand because it's just doing the same thing under the hood, except they've just set this up

[29:07] for you. Um so, we've got this here now. We've got our memory.md, our claw.md. And then now, if we go back into Cawork, if we do a new session in that same folder, and we say, "My favorite color is lavender." Better remember. For the sake of the demo, I hope that it that it does what it's told.

[29:37] It's going to remember it. You see here? Perfect. It's gone good. I'll remember that. Let me save it to memory. And now you've got this big memory file that builds up over time. And whether For example, this is your executive assistant, so it might be saving preferences like how to sign off emails, or don't connect with clients on Slack. I always want to keep client comms on email. But if you're building out like a head of marketing, it might be preferences about how you like your ads structured in Facebook Manager. If you're building If you've got a folder where you're working on a website or an

[30:10] app, it might be things like don't use dark mode, and then it will update so it'll never use dark mode again. And these just compound over time. So, as you start to build up these rules, the amount of errors go down. And And this just compounds and compounds over weeks and months. Remy, have you seen some of these memory.md files get so big that at a

[30:32] certain point it's just ineffective? Great question. I personally haven't happen have that happen to me yet. I haven't hit that threshold. Um but a best practice for those Claude MD files is to keep it around like no more than 200 lines. And yeah, I could imagine if you started to build this up over years and years, you'd eventually hit a point where all the the rules are stepping on each

[30:55] other's toes. And you know, you could probably go through and do a bit of like a manual clear. But, I haven't hit that threshold yet. Cool. So, people don't need to worry about cluttering their memory MD. I wouldn't worry too much. I mean, if it's saving like this like the silliest little things, like the tiniest corrections, you could maybe update that Claude MD to say only save like substantial corrections, you know? And then you can have a bit more control

[31:20] about what it's saving. So, that's probably a yeah, what I would do there. But, once you've set this up, now when you say something like quit writing so formally, it's going to do the task, then update its agents or in this case Claude MD to keep tone casual, never formal. And then now in any new sessions, it's going to keep that preference over time, which is

[31:41] pretty cool. So, now we've got our executive assistant set up with memory and we've given him his role. We now need to connect our tools cuz by default, most of these agent harnesses, they just have web search baked in. But, if you want to actually start linking it up to your tools like Gmail, calendar, and everything else, which is where the real productivity gains are made, you need to

[32:05] do so via what's called MCP. And I actually got Greg, I got this MCP explanation from when you had on Is it Ross Mac? Ross Mac. Yeah. Yeah. So, this he did a great explanation and it just dropped into my head really nicely. And it's basically that before MCPs, your agent or your LLM, in order to speak to tools, it had to kind of learn their language cuz Claude speaks English, Notion speaks Spanish, Gmail French, your browser speaks Japanese, and Slack speaks Chinese. And it was capable of connecting to those tools, but it like required these extensive custom developments that took a long

[32:42] time. But then Anthropic actually created MCP, is that right? Yep, it's right. Yeah. Anthropic built MCP to basically sit as this translator in between your tools so that Claude can still just speak English and your tools can just speak their languages and this MCP speaks every language and then just translates your calls from your agent to the tool and then from the tool back to your agent. So just set a really easy standardized way to connect tools

[33:13] up. And that's what we're going to be using to connect all of the tools to our executive assistant. So if we go back into co-work here, you can see that Claude make it really, really easy to connect up your tools. You can just go to connectors, browse connectors and they've got like hundreds of all the like biggest apps that you

[33:31] probably use. And you can just, you know, add them, sign in, pretty self-explanatory. But I believe Codex would be the exact same. You know, you can go skills or If we go settings, they probably have like uh And then like Manas is the same. For example, if you go into Manas, we can see

[33:52] we can go and connect our tools. Very, very simple process and then same with Perplexity computer. You know, you got your connectors and you can connect all your tools in here. It's just all using that model context protocol, MCP. So I've already, before the episode, gone and connected all of the tools that I use most, like Gmail, Google Calendar, Granola, Notion. They're all set up

[34:15] already as MCPs. And what I'm going to do now is I'm actually going to open up this executive assistant folder in Claude code to sort of demonstrate how these harnesses are all the same and they work off your local files and the real future-proof AI stack is just having those markdown files on your

[34:33] computer. And the reason why I like to work in markdown files is because it's just the easiest sort of format for your LLM, for your agent to actually digest and understand compared to if you were to give it your files as like a docs or a PDF file. So, I like to use Claude Code within

[34:52] Visual Studio Code. Um so, you can see here, I'm just going to It looks very similar to Anti-Gravity. I'm just going to open up our executive assistant folder here. And the way that I see the future of this all going, Greg, is I think that everyone's going to have their what I call an AI OS, like an operating system. And this will just compound over time, like you saw with adding the rules and

[35:14] getting less errors. But, with adding your tools and then skills, which we'll get into, which is basically just training AI on your processes. And I think that everyone's going to have like an AI operating system they work in. And everyone will just have personal agents and agents to manage each department of their company. And people won't actually use these apps anymore. Like, I've connected up Gmail, uh Google Drive, calendar, Granola for my meeting notes, Stripe for payments, Notion for project management. And I don't even enter these tools anymore. I just sit in Claude Code as one central place. And an example here is I sent myself before the episode, I sent myself

[35:51] an email from a fake prospect. And I also entered in Granola a fake meeting with this prospect. So, now I can say things like um summarize my inbox from today. Um you know, so someone might ask like well, how important is that really, you know, like is that such a high value task? Like our What are high value tasks that you're

[36:19] actually getting done here? [sighs and gasps] So, Or maybe Or Or maybe you get a lot of emails, you know? Well, if you know, emails is a big thing if you do get a lot of emails, but just having like all those tools connected in one place and not having to switch and and copy-paste context. So, you'll see an example here, right? So, we've got summarize my inbox from today, which is like one of the most basic agent tasks ever. But, we can see this is one I sent earlier. We've got this one email here like our call today, excited after your call wants next

[36:50] steps. So, I might just say here, um Okay, great. I review my meeting notes with Maltoshi from today and then draft up the email sending the proposal and creating the Stripe payment link and then go into Notion and set up the project. And where this starts to compound even more is when you start to build out skills for each of your processes, because every time I do a process, even like this, manually prompting it, um and I know I'm going to do it again at some point, I'll then just turn that

[37:23] into a skill. Uh and then you eventually end up if you automate like three to five tiny manual processes each week with skills, you eventually end up um automating like your entire life with these agents. Right. So, it's it's not so much in like summarize my my emails where it's super super valuable. It's like that's where the starting point is and then we want to like manipulate it and use it and go deeper and stuff like that. That's when

[37:50] Exactly. these tools really really are valuable. And you can see here it's now connecting all my tools. So, it's going into Grainola and found the the full meeting of what we went through today. It's now going into Stripe to create the product link. Um Yep. And then it's going into Notion to set up the project. And then it will it should create the

[38:13] draft ready for us to go to send out. That this is really like a new way of working. Right? Yeah, it is. It is. It it it it it's so new and I and I even this task, it's really simple, right? Just sending an email based on a call with a proposal link and stuff. But like even if you can just do something like seven times faster without having to go into all these tools, copy the meeting notes into the

[38:38] page to give it context on your meeting. It really starts to compound and you start to fit like a week in a day and then seven weeks in a week. Um and you stack that up over a year and you're going to be miles ahead of everyone else. Uh and when we get into skills, you're going to see how this continues

[38:53] to get even better. But um you can see here it's drafted the email. It's pulled in all these insights from our call in Granola, which is like where I do my meeting notes. And then it's created the Stripe payment link. Here, ready to go. That's cool. And then now I can just go um

[39:15] send this email. And it will use my Gmail integration to to go and send it. Uh and then really cool. It is, hey? I think this is the new way of working. And Cody Schneider, who you had on the pod the other week, I saw a tweet from him and he said that in the future everyone's going to have like an AI operating system like this and you're going to have like the 100X employee because everyone will come into their role with a pre-existing AI operating system and then build out skills for all their manual processes, similar to how I was describing, and just keep building

[39:47] skills each week for anything manual that comes up until eventually their entire life and work life is automated. Great. So, you can see here it's now created the draft here in Gmail ready for us to to go. And if we're happy with it in the platform, I could also just ask Claude to send it there and then. Uh but then what like also gets really cool is I'm going to demonstrate now how I actually build out skills for these processes. So, I know I've talked a lot about skills so far. I want to just give a little

[40:15] overview on what skills actually are. Yep. So, the easiest way to think about skills is SOPs for AI. So, standing operator standing operated Oh my god. Standard operating procedures for AI. The most [laughter] It means once you explain something once, you never have to explain it ever again. An example of this is without skills, if you are creating a proposal for a client and you're sitting in in in your cloud chat or whatever agent harness you're using and you ask it to create this proposal, you're probably going to go back and forth a bunch of times. Remove change the formatting here. Use this color blue for this part. Put the price at the

[40:56] bottom instead of at the top. And eventually, maybe after 15 minutes, half an hour, you land on a proposal that you're really happy with. And you send it. And the next week, you want another proposal written, but unless you're going and finding the same session and working in that same session, it's going to have completely forgotten all of these preferences. And even if you have that memory system set up, these kind of things you don't

[41:20] really want clogging up your memory. They're better off as skills, which is basically it packages up that process into a dot skill file. And in that dot skill file, it's basically just a markdown file that explains the exact process that you went through. So, you could create a proposal skill. And then every time you now you need a proposal written, it just takes that skill, knows exactly what to do, and then you can have that proposal

[41:47] the same way every single time. So, is a skill like a memory file? Like, what's the difference between essentially a memory.md and a skill? Like, is it just is it is it is it is it just like a memory.md file for particular job to be done? pretty much like exactly it. And all of these agent harnesses pretty much now

[42:10] have skills as a feature. So, you can see here like if we go into Codex, for example, they've got skills here. Um same with Claude as well. And these when you're working with these agent harnesses that operate like mostly locally off your computer, um you can see here it actually operates out of this hidden

[42:33] file called a dot Claude folder. Skills, and these are all of the skills that I've created. There's tons. And if we open [clears throat] one up, for example, um like this one here, let's find a good one. For example, I've got this one here for writing viral hooks. And in this skill, we have a dot skill file, which is basically like your memory.md, which explains the exact

[42:57] process for writing viral hooks. And then it's also got packaged in here some references like hook formulas. Okay, so wait. So, how did you create that skill? Okay, so there's two ways that I find useful to create skills is one you can have an idea of a skill you want to create off the bat. So, like viral hooks for example, I had this course on viral hooks, which I transcribed, put it into

[43:25] Claude, and Claude has this by default. It has a skill creator skill added into it. Same with all of the major agent harnesses, they'll have a skill creator skill. So, it's kind of like skill-ception. You use the skill creator skill, and you say, "Hey, take this course on viral hooks and create a viral hook skill." And it can create it like

[43:43] that. That's one way. Uh and then it will package it up nicely with that skill.md. It'll do the whole thing for you. Wait, you you you asked you asked it to take the course? Yeah, I I uploaded the course like the full transcript of the course. Literally. And I said uh yeah, based on this course on viral

[44:03] hooks, build me a viral hook skill. Uh and then I use that for my content team. So, just like we're building the executive assistant folder, I've got a folder called content team, and I've got that Yeah. that uses that skill for me. And the second way to create skills is going through a process manually once

[44:22] with Claude. And then if you know you're going to have to do it again, like that proposal example, you can just say once you've done the task, "Hey, create a skill for what we just did." And it will package up that process you went through. And that's the second main way that you can create

[44:35] skills. So, in your viral hook example, if you go into that folder again, Yeah. so you have a references folder. Yeah. So, that is probably like was that Yeah, let's open it. I'm just curious. So, here it's got a full thing about like So, was this from the course? he is a bit screwed. Yeah, this was basically from like a a

[45:00] course I put into it. And did you ask it to create a references folder? Like how should people think about No, it just did it. It just did it. So, I think what would be great is if we could actually um, demonstrate building a skill live. Let's do it. Um, and so, for example, like this process here, um, I might I could create a skill called like a daily brief skill, you know, that goes through and summarizes like your calendar, your inbox, and your projects in Notion, and plans out your day for you in the morning. And then you can run that on a scheduled task,

[45:35] because uh, a lot of these agent harnesses now are starting to introduce scheduled tasks. So, you can just run it on 9:00 a.m. every morning, "Use my daily brief skill to prepare me for my day." But I think another cool one here, just to show you an example, right, of how my new like how intricate I make these skills, is let's just say for this fictional meeting I had with this person, I might say, um, "Can you draft up an email? I want to refer Maltoshi to my good friend Sebastian, who has an AI automation agency and can help them out better with their

[46:12] needs. And then we can just go um Sebastian's email is And we can just say that, right? And then now it's going to be able to take the notes from Granola, all the contacts, and then draft an email connecting these two, um a prospect with a friend. And you know, I have different like referral things set up like that with people in marketing agencies. And that's just like a little manual process there. Tiny. It

[46:45] maybe takes 15 minutes out of my day. But then I can just go I want you to use your skill creator skill and create a Sebastian refer skill so that whenever I ask you to refer someone to Sebastian you know exactly what to do and you know his email address.

[47:10] And then that'll build out that tiny skill for us, tiny process. But it means I like I know in the future I'm going to have to refer someone to Seb again. And even if this skill now saves me 15 minutes another five or six times they start to compound when you create skills for every single little process

[47:26] in your business. Yeah, I guess it's like we should just be asking ourselves like, you know, in our day-to-day life, like what are all the jobs to be done? Yeah. are all skills that we need? Like what are the repetitive processes or SOPs as you Yeah. talked about. And then just setting up as many as possible.

[47:48] Right? To make our lives easier. Exactly. And just to give you a little demo here. So I sort of I to that folder structure at the start of the video. And this is it here. So we've got workspaces AI with Remy. And for example, I can open up my content team. And within this folder uh this is just like a more elaborate version of our executive assistant, but we've got our Claude.md in here, which explains um you are like the main orchestrator, you have these sub-agents. It's just a more elaborate version of that

[48:23] Claude.md. But I've got a skill within this for like a meta ads analysis. So that was a process, for example, if you're a marketing agency owner, this is probably like the kind of stuff that you can get inspiration from. Um like ads analyzing, you know, taking competitors ads libraries, breaking down all the creatives um and their landing pages. So I built out this ads analyst skill, where I literally just do ads analyst, and then I paste in like the

[48:51] ads library URL like that. And I'll click run, and I did an example yesterday with the Udi, which is a super large e-com brand, and it ran through and basically scraped all of It took screenshots of all the landing pages. It went and scraped all of the ads that they're running, all like 220. It then did a full deep dive here on all

[49:14] the ads. Visual analysis, copy analysis, why did this work, what could be improved. It basically did a breakdown of all the landing pages with screenshots. Hmm. And it did a master report here about everything that's going on. So it did it just did a full breakdown. And that was like a manual process that I would have gone through when I used to run like my marketing agency, and that probably would have taken me like

[49:36] three or four hours. And then I went through to build out this skill, I went through the process once with Claude. Like I started a fresh session, and I was like, all right, go to this ads library URL, scrape this, do this, do this, do this for for two hours. And then after I've done the entire process I just said use your skill creator skill to make a skill for ads analyzing and package up the entire process we just went through as a

[50:03] skill. And then now whenever I want to do that process again I can just invoke the skill and it and it knows what to do. Which is pretty cool. Crazy. Crazy. Absolutely crazy. So if So now the the refer Sebastian skill is live and now whenever I want to refer someone to Sebastian again I can just say um yeah refer to Sebastian and it will just

[50:25] start to use it will use that skill. That's example there tiniest process but you build like those up for all these little tasks you do day to day and then it just compounds and compounds and compounds. And I can already think of an idea here where you could then you can chain skills together. So you could example have like a a meeting prep skill that prepares you for a meeting by researching the guest

[50:51] um and compiling some talking points. You might have like a podcast research skill for example Greg for a guest that's coming on. And you might also create a a morning brief skill. And in the morning brief skill you can say if there's any meetings uh coming up or podcasts in my day use the podcast research skill um to research the the the guest. And you can like chain them together uh and build like some really really

[51:18] cool workflows. Yeah and you can have it so it sends you an email right? Yeah exactly. And then now these harnesses are starting to get more and more autonomous. Like they're starting to add like you know in the car they're starting to add like cruise control and stuff. Like now within most of these harnesses you can schedule tasks um like in co-work or cloud code now and you can like for example this one here you can go new task and I could say like uh

[51:46] run my morning briefing skill. And then set that to go every every morning at 9:00 a.m. And then now it's like an automated workflow that like you've just got running every morning now, which is pretty cool. Yeah, I'm I'm doing this right now. Like for example, I'm I'm buying a new car right now and it's like a particularly unique like color that I want and feature set and there's just none none really available, so you know, every 3 hours I I have I'm scraping all the different car marketplaces

[52:23] That's great. and and then I'm getting a notification that you know, it when something comes up and it's it's crazy, right? Like it saves me I'm one of those people that like if I didn't have this, I would be spending an hour of my day just like checking religiously every single, you know, CarMax and cars.com and Autotrader and all these websites and refreshing like a

[52:49] insane person. Um so yeah, the schedule to not great great example there. But you know, this like this is a skill that would be relevant for my executive assistant. Same with that car one. That could be a good executive assistant skill. But then I've got those more elaborate skills built out for like, you know, my um content team that ads library scraping one. And then um I've got like, you know, re- a research weekly research skill for my newsletter team. And that runs on a schedule every Thursday morning to go and scrape like I've built the skill out so it goes and scrapes Twitter and Reddit to find what's new in

[53:23] AI. Um but yeah, skills are so so powerful. Um combine them with like your MCP so it can use your tools and then you can start to just train up your agent on all the processes in your business. And I did a um a build out on Open Claw for a agent to manage meta ads, and it went

[53:43] pretty viral. And the the way I built this was with all these key concepts. So, Open Claw functions the exact same way. So, I I hope it hasn't timed out, but I've remote accessed into my Open Claw dashboard here. And you can see it's just operating off an agents.md file in the backend. But, instead of the dot claw folder, it's in a dot open claw folder. And then it's got a couple of these other ones here. It's got a

[54:07] memory.md. It's got some of these other ones that it's added on, like a soul which tells it its personality, and an identity which tells it who it is. But, it's that same concept of markdown context files connecting your tools, and then creating skills. So, that uh meta ads manager one that went pretty viral, I just planned it out with Claude. I was like, "I want to have this Open Claw

[54:31] manage my meta ads. Help me write the agents.md file to tell it you are my meta ads media buyer. You do these processes." And then I created skills. So, I created a ad creative skill. Um so, it knew to go look in the Dropbox folder and create creatives. I created a copywriting skill, so it knew how to write good copy for the

[54:55] business. Um and I just built out there's probably maybe 15 different skills. And then I would combine scheduled tasks, which is cron jobs, with skills, and the context files, and then just give it all the tools it needed. Um and just following the same process as we just went through to build the executive assistant, I had an Open Claw

[55:16] meta ads media buyer, which was sick. I love it. And for the beginner, are you like would you recommend people, you know, use Open Claw, or should they be using Co-work or Manas and some of the ones you showed? So, great question. Uh I would say that Open Claw is probably like one of the hardest to learn and set up of these harnesses. I would say

[55:39] Claude Code is probably the easiest. I think Perplexity Compute, you did a video on it. Um it's pretty simple, too. Same with Mana. Easy. Um but I would definitely learn um and get comfortable using like Claude Code or um one of these other ones before I started to play around with Open Claude. And I would also uh have all the processes built out in Claude Code first. So, for example, that executive assistant, over the next uh 2 weeks, I might build out a bunch of skills, like the Sebastian refers skill, um like a daily brief, meeting prep, etc. etc. And then once I'm happy with how it's all functioning in Claude Code,

[56:17] then I could look to migrate that into Open Claude, where it has that more autonomous nature to it. So, that's kind of how I think about using Open Claude and those other harnesses. Yeah. Cool. All right, anything else you wanted to Um cover? I mean, like really, there's no right or wrong way to run these. Like that was

[56:35] the executive assistant. I've got one built out for all the other departments in my business, and then other businesses I work on, I have the same. Um and you can just kind of build out that structure with what works for you. Um you've got like one other thing to mention is global versus project level, which I'll just go over super quick. So, like those skills, for example, um you can add them at a global level, which means they apply to every single project you work in, whether it's the executive assistant, your head of marketing. And some skills you want globally cuz you might use them in every

[57:08] chat. Um like a um truncate skill that I created, which just makes whenever I want to make something shorter, it makes it shorter without compressing the sentences, but just removing sentences that don't need to be there. And that's something I want in every session. So, I've got that in global. But you can have project level skills, like that um Sebastian refers skill, I would not want that with my marketing head of marketing cuz it's just like plugs up the context

[57:38] and you don't need it there. Um, so I would have that as a project level for example. And you can have um, global skills versus project skills, global Claude dot MD versus project Claude dot MD. And same with MCPs, you have global MCPs and project MCPs. That's probably the other concept um, to go over. But look, other than that, that's pretty much the entire agents crash course. So, it's just that loop running in the back end to complete your task and connecting in your tools, your

[58:07] context, and the LLM all in one place. Uh, and I would just say to to work out what roles you want to start to build out an agent for, go into Claude or your favorite chat model and get it to help you build out those uh, context files through an interview style process. Just say ask me questions

[58:25] to build this out. I would connect all the tools that you need and then start building out the skills through daily use. And then pretty soon you're going to have like pretty powerful AI agents built for every single um, aspect and department of your business. Ramy, thank you so much. I'll include links uh, in the show notes, in the description, where you can go follow him, get to know him a little bit better. And uh, I appreciate you coming

[58:50] on dropping some sauce. Thank you, man. Thank you so much for having me on, Greg. It's been a blast.

---

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