Grok Bot Clearly Explained (& how to use it for beginners)

0:00GrokBot has just planted the seed that will change the way we interact with computers forever. It hasn't done anything crazy. There's no technological breakthrough. They've just taken a few features, a few ideas, and they've combined them into one package that I think will represent sort of the next ChatGPT moment in the world of AI. Now, what do I mean by a ChatGPT

0:31moment? Let's imagine it's early 2022. So, ChatGPT has not come out yet. What OpenAI do have is a completion model. It's publicly available, but not that many people know about it. And what it does is you can feed in a piece of text like this, and what it will do is it will complete the text. It will complete what it thinks will come next. So, in this example, it's basically like a series of sort of instructions that you could imagine sort of being given to a teacher in a sort of a lesson planning sheet or something like that. So, here's the completion model in action. You can see

1:09that there's an initial prompt going on, and then the model in green is going to fill in the rest. Now, outside of the software developer community, not that many people know that this exists. And on its own, it's the use cases are kind of ambiguous, right? But OpenAI did something really special in the latter half of that year, which we all know about, which they created something called an instruct model, which basically became ChatGPT. And in it, the instruct model, what the researchers at OpenAI did was they trained the model to take in a query like this, or take in a prompt like this, and respond as if it

1:50was a question. So, the response then, what the completion of this piece of text was, was the answer. And of course, what they ended up doing was wrapping that inside of a consumer interface to put out on the internet and share with everybody. And we all know the story. It was quite It was quite revolutionary. Now, the point I want to make here is that the technology underpinning ChatGPT, it already existed months before ChatGPT did. It's only that what OpenAI did was it wrapped it in a user interface that made it palatable to the masses. And I think this is what Grokbot

2:29is doing for AI agents. Now, most of the world is still stuck using these AI models like a glorified Google. They ask a question, put in a prompt, and get an answer. And then they go to the next chat, maybe the next day or the next hour, and start a new one. But the shift that is coming is using these models not as glorified search engines, but as agents, as intelligent digital beings that can do actual work, useful work on our behalf. We can hand stuff off to them like we would an employee. Now, this idea of AI agents or

3:10agentic workflows is nothing new. They've been around for a few years now. It's just that they haven't become mainstream. It's a small portion of sort of knowledge workers and people who work in tech who are making use of AI in this way. And it's because the user experience around using these agents is just not up to scratch. We're still stuck in this Google search way of thinking in the way that we interact with them. And that's been limiting the ability of the public at large to use AI in this sort of agentic way. There's also the problem that like if we take Claude and they've got agentic products,

3:49Claude Code and Claude Co-work, they largely or the default is that they operate while your computer is open. So, those agents, they'll start up on your computer, and then there'll be a server somewhere where the AI agent will actually run. And so, your computer, via you, will basically tell that server to start doing some work, right? Summarizing some emails or what have you. And that server will go out on the internet and it will

4:16do whatever work that it needs to do. And when it's finished, it will return the result back to your computer. There's just one problem, which is that if you close your computer, then the server here, the AI agent, can't return the work back to you. So, part of the user experience that Anthropic, the makers of Claude, as well as OpenAI and Google and all the AI providers, part of what they've done here is they've let their AI agents actually operate in the cloud by default. So, your computer just tells an instruction to the server for the AI agent to do work. And so, it the server spawns an agent, right? This

4:57agent just it is created at that moment when it is needed. And then, if you can give it, you know, access to certain accounts, like your emails or your project management software, and you give it some instructions, role, how it should behave, then it can go and do useful work. And then, simply, when you're when you open your computer, you can see the results of that work cuz it doesn't live on your computer, it lives

5:22on the server. The problem there is that because these agents are temporary, they're just spun up on demand, this environment setup, like given it the set of rules, the set of instructions, given it access to different accounts, like that is a place where things can break. That's a surface area of the agent where things can break and requires often just a little bit of technical faffing about, to put it frankly to get that to work reliably. Now, there have been some wonderful innovations to solve this problem, notably Open Claw, which gives an AI agent basically a permanent home. So, you always have this agent running. It doesn't get spun up on

6:06demand, and that means that you can give it access to different accounts, and you can give it certain instructions. You can access to the internet, and it can just always be on, and that means that from any device that you own, right? You can simply connect to the internet from either of those devices and send the request through the internet to your

6:35your AI agent, your always-on AI agent. And that's They're often run on these computers that just live at your home that you connect to the internet. So, this is a Mac mini, very popular computer for running these these Open Claw agents. The problem is that just to set this up is just like a very technical. It's great conceptually, right? An always-on agent that you can just talk to from anywhere, right? That has its own personality, its own set of instructions, and can run things on a routine. Like, that feels like the right way to go directionally. It's just that this has been technical to set up. And

7:12this is where the story turns to Grok because Grok Bot just basically turns this very technical setup that people have been doing for something like Open Claw into a one-click process. You sign up to an account, and what you get is an AI agent, a Grok Bot. Essentially, you actually get multiple of these on one account. And they all live on a computer on the internet, right? On some server,

7:42somewhere. And they're always on. Which means that from your devices, you can ask them questions, delegate work, and they can go off and do that work all on their own. You'll give each of them their own personality, their own set of instructions. And I'll go through some examples here in a minute, but you can imagine here they've all got, you know, slightly different set of instructions, and they'll just go off and do their work, right? When they're finished, they'll tell you, and you can check on in on them as if you were messaging somebody on your team, as if you were messaging a colleague. So, these are truly autonomous AI agents,

8:28which is why I think we're at this ChatGPT equivalent moment for AI agents, because we're going to move out of this world of people just inputting queries one by one as if it was Google search, and they're going to start treating their AIs more like actual human beings with persistent, durable personalities and memory that they can talk to, that they can collaborate with, that they can delegate work to. So, I want to show you some examples here. So, I've got a fresh Grok Bot account set up here, and the onboarding kind of shows you how it works. So, each Grok Bot has its own computer, just like you do, somewhere in

9:09the cloud, where so it can do work autonomously. We're going to give each of our bots a particular sort of personality, and it's going to have its own memory and set of instructions about how to do its work. Now, you have this option in onboarding to connect some tools. I'm going to skip this, and we're going to do it later on from the interface. So, let's start with a really simple example. We're going to have a finance bot. This could be a bot just for your personal finances, for your business finances. Let's make it gold because it's, you know, it's finance related. And we can just here explain to

9:43the bot in plain language what we want it to do. So, the first thing I want to do, I want it to I want to give it access to my bank feed and I want it to tell me about all the recurring subscriptions that I have. And I want it to basically check in with me periodically to say, "Hey, do you actually need these subscriptions or can we cancel them?" Now, in the interface here, you've got two sides. You've got here the actual computer that this bot is running on, right? So, it has its own computer, its own Google Chrome that it can open. You can actually click around

10:14in here and do stuff on their computer. This is the way that you take control from the bot when you need to, for example, log in to certain accounts on its behalf. But down here, you simply explain to the bot what you need. And this is usually always the best place to start. You've also got a little microphone here, so I can just explain to it, "I want you to check my bank account every month, specifically my transactions, and I want you to tell me what recurring subscriptions that I have so you can check in with me about which ones we may want to cancel." So, you can

10:51see here that I've told that I'm with Wise Bank and it's actually opened up its web browser. It's navigated to the Wise website. It's actually asking me to log in here. [snorts] Now, one tip, wherever possible, let's avoid doing this. Avoid giving bots access directly to your web accounts via the interface, just because that then gives them all the permissions that you can do as a user. It's also slower. They have to kind of scroll through the web page. It's slower, that is, than if we give them direct access to the database, so to speak, behind these accounts. So, in technical terms, that means giving them API access. So, that's just a

11:34backdoor access that we can give AI agents like this to our data. Now, you can always just ask the AI agents, "Hey, is there an API for this tool?" I know that for Wise that there is, so I'm going to say, "Let's use the Wise API." You can also see, just while it's working here, it's detected that

11:55this needs to be set up as a routine. So, something that it's going to do on a schedule. So, it's said here, "Hey, I'm going to do this work on the 1st of every month at this time." You can also always come in and create your own routines here by describing what should happen and adding a trigger for on a schedule. There's other triggers here, as well. I'm going to show you an example of some of these later on. Okay, so I have followed the instructions to get an API token. If you're not super familiar with APIs, this is just a password that allows one system like

12:30Grok Bot to access the data in another system like my bank. So, I'm going to paste the account here. Whenever you're adding credentials to Grok Bot, make sure that it is giving you a box like this rather than just pasting the information here in the chat. That's because these chat chan- transcripts will be on Grok's servers. They might get on to certain logs where the security of those logs is less than if you put it into this field here where it will be scoped only to this Grok's this Grok Bot's computer on the cloud. It's its own secure

13:08computer. So, I'm going to save that, and it's now going to continue with the work that it needs to do in order to pull my bank account transactions. Now, while this is working, I'll show you another example of where these agents can be really useful, which is just periodically checking certain websites to see if anything changes. This could be used in a load of different circumstances. You can imagine here that you'd want to be notified whenever a competitor's pricing changes,

13:36for example. So, this is something which I'm actually going to delegate to a new bot that I basically just want this to be in my marketing bot. So, I am going to click on it here and just call it my marketing bot. I'm not going to over engineer anything else here. So, it's going to be my marketing bot and I just want to say that I want you to watch this website and tell me when the pricing

14:05changes. And what it will probably do here is it will on some kind of recurring schedule, it will check this page. So, you can see here that it will check hourly on weekdays. Now, this website is actually one that I control just for demonstration purposes. So, I'm going to update the pricing here and we'll see if this bot picks up on it. And also just for the sake of demonstration, I'm going to make this check every minute, um so that we don't have to wait too long for this this routine to kick

14:41in again. And while it does that, we'll go back to the finance bot. And it's given me here an overview of some of the subscriptions that we have. If I was feeling like it, I could actually get it to cancel some of these subscriptions itself. It would probably open up a web browser, ask me to log in, and then it would go through the steps needed to

15:05cancel those subscriptions. Now, while we're waiting for our marketing bot to check the website, let's create another bot here, which is a very common type of bot that you're probably going to want to create, which is just an executive assistant, which will do a lot of different important jobs for you. And what I want this bot to do is every morning just prepare me for any meetings that I have that day. So, every morning at 8:00 a.m. I want you to check my calendar, any meetings that I have that day, I want you to pull the relevant context for those meetings and give me a briefing. What do we need

15:46to discuss on that meeting? What are the action items that we need to get out of those meetings? And what's any relevant context that I might need going into that meeting or any to do items that I need to do ahead of that meeting. Keep the briefing concise and I'll ask any follow-up questions if

16:06I need to. Now, for this agent to do its work effectively, I obviously have to give it access to my calendar and also give it access to any places where it might need to get that project context. So, I'm also going to tell it here the context for projects for where I will have client meetings will live in Fireflies and the meeting recordings and in GitHub in various repositories that we have for these

16:36client projects as well as ClickUp. So, I'll tell it that and it should then give me the instructions that I need in order to connect to these various tools. And all of these connections will exist on this shared computer that all of these bots have access to. So, all of the bots have access to the same data. It's just that they'll have certain memory, certain roles that are scoped

16:59just for them. So, I'm going to tell it here, we need to connect Google Calendar, we need to connect Fireflies, ClickUp, and also GitHub. Now, a lot of these are plugins that are natively supported by Grok Bot. So, under this plugins tab here, there are a lot of plugins that you can just connect to with one click. Other ones like that Wise bank account that I've shown you, you may need to integrate by

17:23pasting in an API key. So, because ClickUp and Google Calendar have these official plugins, I'm just going to go and authorize them here. And what that did is just prompted me to log in to ClickUp. So, same thing for Google Calendar. Okay, so I've been going through the rest of the instructions for GitHub specifically. Again, this is something that we've got to look out for them is that it just asked me to paste

17:46my access token here in the chat. Remember, that is a no-no. So, I wanted to give me the secure input for pasting the token. Remember, at the end of the day, these are AI agents. So, they do make mistakes. They're not predictable machines. And they're generally no more reliable at following instructions than human beings are. So, sometimes we have to

18:14remind them to do these sorts of things. So, here is the place where I can paste my token. And while this is completing the remaining setup, it looks like our marketing bot has a notification for us. And by the way, I can access this grok bot app on my phone as well. So, I can get push notifications. And it's telling me, "Okay, looks like the pricing has been updated." So, if I go and I check this myself,

18:41indeed, there is the new rate. So, it looks like the morning briefing is set up. I actually want to test this out. I actually had a lot more meetings yesterday, not today. That's why I'm recording the video today. So, I'm going to say, "Create a morning briefing bot as a test run for yesterday, Tuesday, the 8th of September, as if it ran yesterday at 8:00 a.m. before my meetings for that day." So, here is the test run. I blurred it out here cuz these are real clients. But, if I wasn't happy with this, what I could do is I could tell this agent to update its instructions

19:18for how to generate these briefings. So, inside this morning briefing, I have some instructions here about how concise or detailed that these reports should be. I could put in here more detailed instructions. I could even provide an example report to the bot, which it would then store on its computer, right? Like, as a file on its computer that it could then reference before generating these daily briefings. So, this is really just scratching the surface just

19:48to show you in principle how it works. But, just in this chat here, you can go back and forth to make sure that the bot is remembering your preferences for how you want these reports to look in the future. And that goes for any assets, anything that these agents do. Remember that because it has its own computer open, you can save whatever you want here, right? You can save an example spreadsheet format that you want to use as a template whenever it creates, I don't know, a new model to send to a client, right? Create files on your computer, documents on your computer that are style guides for how

20:25to send emails or how to write messages to customers using the company style guidance, right? Branding information for creating assets. Like, anything that you would keep in a shared repository for your team to access, you can put here on the computer for your Glock bot. And I want to show you one last really cool example here. I'm going to create a new bot here called a project manager bot. And this is a bot that, like my executive assistant, can get context from all of my projects, but explicitly I want to keep the scope to like doing

21:01stuff on projects, not company-wide. And the reason why is cuz I actually want to allow this to be invoked or triggered from Slack messages from our team Slack channel, so that within our Slack channel, our team can actually talk to this bot, and this bot can go out and do its work and report back in the Slack channel as if it was another

21:23teammate. So, I'm going to describe those instructions here to it. I want you using project context that we already have installed on this computer, emails, Fireflies, GitHub repos, ClickUp, I want you to answer questions about projects, and I want you to set yourself up to be invoked or triggered from within Slack. If somebody at messages you, I want you to be able to respond in Slack with an answer to their query about a particular project. Okay, so the agent went through the process of getting me connected to Slack. In the process, I encountered every YouTuber's nightmare when doing these sorts of videos, which is just a

22:06like a really annoying technical hiccup. Thankfully, we're in the age of AI agents where you can basically have a chat with it, and it can help you to work out what the problem is. The solution in this case is that we just have to in Slack use this @PM tag, and I've actually done that here in one of the project channels by at mentioning PM and then asking a question that's gone straight to our Grok agent, which of course now has access itself to all of these different tools and context, like our ClickUp and our Fireflies recordings, and so it can give me a contextual answer directly here within

22:49Slack. So, this is a type of routine in GrokBot that is running on a native Slack trigger. You can also set up triggers here on something called webhooks, which is this is just like a universal way that any system can connect into this bot. So, you could actually connect this up to an email address that maybe you give to your clients so that they can just ask a message to a basically a support email address and get a reply directly to that email address that is written by that one of these Grok bots using all of the project context. They might be asking for an update on certain tickets in

23:29ClickUp. So, not all clients, of course, are going to be as technically savvy as others and so being able to provide answers where they're already comfortable, i.e., their email box, can be really handy. There's actually a really cool service called Agent Mail, which is set up for exactly this use case and I've been wanting to play around with it. So, if you want to see me do a video about that setup with Agent Mail, just let me know in the

23:51comments. Okay, so this is obviously just scratching the surface of what Grok bot can do and I suspect that all the main AI players like Anthropic and OpenAI, they're going to adopt this kind of user experience pattern of having sort of a sidebar of agents that you can talk to because it's just very intuitive. You can talk to them from your phone, you can talk to them from your computer,

24:14wherever they are and they're always on. Now, if you're interested more broadly about how we as an software development agency working with clients every day on a multitude of projects are using AI in our day-to-day workflows, check out this video up over here. Otherwise, I will see you in the next video.