# Grok Bot Tutorial for Beginners (Automate Anything with AI Agents)

Channel: Zinho Automates
Video: https://www.youtube.com/watch?v=_JYh3tgJzVw
Duration: 15 min
Language: English
Words: 2705
Transcript page: https://viewrankai.com/tools/youtube-transcript/_JYh3tgJzVw

---

[0:00] You've probably seen Grok Bot everywhere the last 2 weeks and you probably scrolled past it because it used to be about $200, but now it's only $20. But let's talk about what it actually does and how you can optimize it because most people still use it like a typical AI chatbot or even an AI agent, but it's actually so different. It doesn't just answer questions or do tasks with connected apps. It can click through with the old websites on its own computer which really means that it can use software with zero integration. So no API keys are even needed. There's no plugins. There's none of that. And the

[0:43] best part about it is that it keeps working after you close your laptop. So when you wake up to grab a cup of coffee, your research is already done. Your inbox is already sorted and the people that you needed to reach out to have already been messaged. Not because you built some complicated automation, because you told your AI agent what you wanted and you only did that once. So I'm going to be showing you how to set this up from zero and then four use cases that you can copy today and the pattern behind all of them. So by the end of this video you can build one for

[1:21] whatever your actual work is. Now if you want the exact terms, all of it is inside our free school community and the link is down in the description below, but right now let's get started and let's hop over to the laptop. Okay, so first things first, let's get this thing installed. Okay, so Grok Bot

[1:41] has a desktop app for Mac and Windows. So once it's installed, you are going to sign in with your cursor account, okay? And this is important to know, okay? So Grok Bot isn't its own separate sign up. It rides along with your existing cursor or super Grok plan. So all you have to do is just sign in and let's get moving. Okay. So, now we are inside. Let's create our first bot, okay? So, pay close attention here because how you set this up is so important. Now, I'm just going to give it a quick short name. So, let's call it something like let's call it scout. Okay? Then, you basically

[2:18] don't need to give it a description and it's going to be the primary thing that this bot does. Now, lastly, let's give it a label. So, if it's primarily a research helper, then label it research, okay? It's going to make things so easy. So, now it's time to give it an actual job. So, I'm going to tell it to monitor my email inbox and to help me triage every morning. Okay, got that. So, here's where beginners usually mess this up, okay? They type in something that's pretty vague, something kind of like help me with email. And then, they basically sit back and they wonder why the output is so generic. So, instead,

[2:52] I'm going to give it a task using a specific structure. So, this structure is the difference between a bot that works and a bot that somebody just guesses. So, you are going to need five parts in your plan, okay? That is the outcome. So, that's what you actually want at the end. And then, the sources. So, where where it should actually be pulling from, the constraints. That is what it shouldn't be doing, the deliverable, which is what the final outcome should basically look like, and then the one that everyone skips and honestly, it's the most important one, the review point. So, where it stops and it waits for you before doing anything

[3:30] that can't necessarily be undone. Now, that is outcome, source, constraint, deliverable, and review point, okay? So, please remember those five and write it down, keep it somewhere safe because every use case in this video is basically built on them. Okay. And now, look at this panel over here, okay? So, that is your bot screen. Okay, cool. So, now let me just open that up real Okay, so that's a real computer with Google Chrome, file manager, and terminal. And it's not running in a chat window. It's actually using an actual desktop in the cloud. So, when it researches something, it's opening a

[4:04] browser and going site [music] to site. So, basically it's doing the same thing that you would on your own. And since that machine isn't your laptop, you can close yours and it will literally still keep on going. Now, once the draft comes back, I review them and then I'll just fix like a couple of things and then here's the step that most tutorials

[4:25] about this topic usually skips. I actually save this exact process as a skill, so it follows this exact method the next time I have a similar task. So, now that you've seen it work, we need to understand its fundamentals, okay? So, there's four things that you need to know before you build your own team like this. Number one, bots are employees, not chatbots. And that's exactly why I gave Scout just one job instead of a list of 10 or 20 different jobs. The second you make one bot do everything, the quality almost always drops. Now, number two is your account gets one cloud computer. Every bot then

[5:05] shares it. So, that means that there's not one machine per bot. The upside is logins sign into Gmail once and then every bot after it inherits the session. Now, number three, and I must be honest, this is a big one, it clicks, it doesn't call APIs. So, that's basically the whole trick because it drives a real browser like basically the same way that a human does. It can operate things with zero integration, so WhatsApp, a supplier portal from 2009, a CRM older than some of you even watching this

[5:35] video right now. If a human can click it, then that means that this can click it, too. Now, number four, and that is bots delegate to each other based on their name, title, and description. So, that's what turns them into a team, and that's the last thing that I'll show you because it only really makes sense once you've got a few things done in it. Okay? So, stay tuned for that. Now, the last thing that you need to know are skills. And you don't need to write a skill. You simply just do the task once, then it it records what you did, and then it becomes the

[6:08] skill. So, that means you demonstrated instead of having to describe it. And if you put that skill on a schedule or trigger like a new Slack message, and it's a routine, then that's the moment a one-off task becomes automation that's now running in the background. Okay, so now for the fun part, let's hit these four use cases, and they are all going to build on top of each other. So, use case number one, that is the the research bot. So, this is the one that I'd start everybody on because it's the safest, it only reads, it can't break anything. So, you create a bot, you name

[6:46] it, and then you give it one job. Then you go find out some things for me, basically something like that. Then you can hand it a real task with those five parts that we covered earlier on in the video, and you can just say something like research these five companies, pull the pricing, their positioning, and anything that they've announced in the last 90 days. Okay, use their websites and public sources only. Give me a one-page comparison with a link next to every claim. And don't guess at anything

[7:15] that you can't find, mark it as unknown. Simple like that. But as you can see, I'm hitting every single point. And now, watch what it actually does. It opens up a real browser, it goes from site to site. It's not pulling from a database that it it was trained on months ago. It's a looking at the page the same way that you would, and it's doing

[7:35] everything right now in front of us. Now, that's the read-only version, okay? So, now let's put it on a timer. So, for use case number two, it's the daily summary bot. [music] So, it's basically the same idea, but this one runs without you. So, you basically just point your bot at the places where most

[7:58] of the time is taken away in your day. Your inbox, your calendar, your Slack, whatever you use, [music] and then the job that you give isn't summarize everything because that's basically how you end up with a wall of text that you are just going to skip over. The job is, tell me only what changed since yesterday that actually matters to me. And for [music] each thing, tell me where it came from, why it matters, and what you think that I should do about it. Then, [music] take that working task and turn it into a routine so that it fires every morning at 6:00 before you even wake up. And I mean, by the time

[8:35] you sit down, the brief is waiting for you. And here's the part that makes this even better. When it gives you something useless, you simply just tell it that all that all that it gave you was noise. When it surfaces something good, you tell it that that that it gave you was now useful. So, then that means it's going to start tuning to what your preferences are. Both of these bots have only been reading so far. So, now, let's give one the ability to actually do something for

[9:05] us. That takes us over to use case number three, and that is the outreach [music] bot. So, this is where it stops being a research tool, and it starts becoming a true worker. The job here is, find somebody old, people, figure out something to say to each of them, and write the message. So, you'd basically [music] tell it, go through the list, research each one, find a genuine reason that they'd care about what I do, and

[9:31] draft a personalized message for each. Skip anyone that I've already contacted, put everything in a review list, and do not send anything, okay? Now, that last part is not optional. Every outreach bot that you've built ends at a research point until you've watched it work enough times to truly [music] trust it. So, always remember, read, research, draft, and then stop. And this is where

[9:57] foundation point three really pays off. Outreach doesn't live in one clean place. It's on a messaging platform or a CRM with no integration or some portal that's never had an API in its life, okay? So, a tool that needs a plugin for everything just stops at the edge of what it's connected to. Now, this one doesn't. Why? Because it's actually doing the clicking. So, once the drafts are consistently good, then what you want to do is you want to save it as a skill, you want to put it on the routine, and now it's a research and drafting overnight, and you are simply just approving in the morning, which is

[10:34] going to save you a bunch of time. So, now that is three bots, each one doing one job, and here's what happens when you stop treating them as separate. That's going to take us on to use case number four, and that is the bot team. So, remember foundation point four that we mentioned? Bots delegate based on their name, title, and description. So, take the three that you already built, the research, the summary bot, and the outreach bot. Now, build a fourth one and call it something like chief of

[11:07] staff, okay? Sounds pretty professional. And then what you can do is you're going to give it one job, and that's going to be to coordinate the others. So, I'm pretty sure you can see where I'm going with this. You stop assigning tasks to individual bots, you go to the chief of staff, and then you simply just say, "I need a campaign for this new segment by Friday." And it hands the account research to the researcher, the drafting to the outreach bot, and then it comes back to you with the finished package, and you didn't do the routing, and it reads the rules, and then it figures it

[11:41] out for you. This is the actual reframe of this whole video. You are not building an assistant. You are literally building an entire org chart, and every bot that you add doesn't just add one task. It adds something the rest of your team can now handle work to. Now, those four are examples. They are not a menu, okay? Because here's what I actually want you to take away. Nothing about those bots is special. There's no template, there's no marketplace, no pre-built thing that I downloaded. All four are the same three ingredients: a name, a job, and a task that then with those five parts that we mentioned in

[12:25] the beginning of the video. And that's basically it. That's the entire product. So, don't copy my use cases, copy my pattern. Ask yourself three questions, okay? What do I do that's repetitive? What do I do that's mostly a reading and gathering before any real decision gets made? And what do I do at the same time every single week? So, whatever you come

[12:51] up with, that's your first bot, okay? Property listings, client reporting, restocking inventory, competitive pricing, grading, invoice chasing, following up on quotes. It genuinely does it not matter what industry you are in because it isn't limited to tools that agreed to integrate with it. So, if the job happens on a screen and a human could click through it, then why not

[13:17] give it over and let this do it for you? Now, real quick, I want to be honest about where this still falls short, okay? Because I'm not going to pretend like it's it's totally perfect. This is still early beta. So, there's a no model picker, so you can't steer towards a cheaper model for simpler tasks, and heavy use can and can't pass your plan's allowance. That shared computer that we spoke about it basically means that your bots aren't fully isolated from each other. And it's strong at continuous multi-step work, but it still stumbles on subtle judgment calls a human would usually catch instantly. So, here's where you actually are right now. You

[13:58] know how to install it. You know how to name a bot. You know how to write a task in five parts, how the cloud computer works, and how to turn something that it worked once into something that runs every morning without you. So, pick the most boring, most repetitive thing on your plate this week, and then build one bot for it. All you have to do is literally just build one. Then, make it read-only, and see what it comes back with. Now, everything that I showed you, the five-part task structure, the full setup, the problems and all, all of it is inside our free school community. The

[14:30] link to that is down in the description below, and I need to mention this, it is completely free. And there's over 34,000 people currently in there having a bunch of fun, and I know you are going to love it, too. Now, if this helped you out, then hit that like button, subscribe to the channel, and turn on this bell notification so that you never miss another upload. And also, check out the video popping up on the screen that I have now, and check out some of the other videos on the channel. And as

[14:52] always, I'll catch you on the next one.

---

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