# I Built a Grok Bot team that organizes Itself

Channel: M Two M
Video: https://www.youtube.com/watch?v=ACar5zEx43M
Duration: 11 min
Language: English (United States)
Words: 1857
Transcript page: https://viewrankai.com/tools/youtube-transcript/ACar5zEx43M

---

[0:00] I built my bot army this morning. They share one computer in the cloud. They communicate with each other, and I can talk to them in a WhatsApp, Telegram-style chat interface. This is Grok Bot. Each bot or agent has a specific purpose. So you provide a name, a title, and a description. What this agent is for. Researchy is my research assistant for deep research, clear summaries, and

[0:24] reliable sources. So now the first research started, and I can continue directly in this chat. There are no separate chat windows. I can open a new thread in here, one single thread which contains the context of the whole conversation. But typically, we stay in the single chat window, and Groqbot manages all the context for us. I can also quote the entire response here,

[0:49] or I can just instruct Researchy. And whilst it is working on this task, I can add the next task. Now Researchy works on both of those tasks. I can still talk to it in the meanwhile, and we saw that we are orchestrating the tasks. So we are first working on the Udemy topic, and then we'll dive deeper into

[1:10] SpaceX. All this happens automatically. And when I ask it what it's working on currently, it also gives me an overview of what is happening. So this is the core idea behind such a bot, such an agent, one chat window tasks. In parallel, always able to talk to you and not forgetting any context. The whole context management thing is done

[1:32] automatically by Grokbot. Now we can leave the single chatbot logic and also explore the computer the chatbot has. On this machine, the computer has access to the browser, the file explorer, and also the terminal. So, if we look at the file manager, we find a reports folder with reports that were generated earlier in Grokbot by

[1:54] myself. And if we check the browser, Grokbot can use the browser as we do it. So, we can instruct it to browse websites. And to do this, we'll first add two more bots. Data-wise for my data analysis and reportify my report generation bot, and it's always important to be precise about the title and the description. This is basically the system prompt that you provide for this bot to instruct him on what his general purpose, what his

[2:24] general task. Actually is, and here I am a bit specific about what I want to have. I want my reports to be self-contained HTML pages and host them publicly on here dot now. Now the HTML page logic is simply using the concept behind a website to render the output because I personally like that and

[2:46] I think it's. And here, dot now is a plugin that allows us to host those sites easily. Plugins are connectors, so to say, additional tools we can connect Claude bot to. In our plugins in the marketplace, here. dot now is listed as a plugin, so Grokbot is able to install those plugins then to

[3:09] run this task successfully. So now I want to dive into Ferrari's recent release, the Luce. If you like cars, you might have heard about that. And I also want to include data wise and Reportify for the analysis and the report. Now it's the point where those bots, those agents, communicate with each other. They all share the same computer, so it's one computer shared by any bot you are

[3:33] created, and they know about each other. Therefore, we can orchestrate the tasks now in this initial research bot, and it then delegates the tasks and makes sure that all tasks are completed successfully. Whilst this is working, we can again look at plugins. Because we cannot only use Grokbot from the desktop, we can also use the mobile application, which we'll explore a bit

[3:55] later. But we can also use it from other applications. In my case, I will use Slack here. I added this to my machine already, so you have to add the account here and authenticate Slack, and then you can add tools like this to specific bots as a so-called routine. Routines are always here in the right

[4:16] section of each bot. So if we go to research, once again. Create the routine Slack research. So I want to initiate research according to the message sent to the Slack channel. So for each new message in my channel that contains the word research, I want this research bot to start with the corresponding research.

[4:37] So now this is set up. I also added a message to my Slack channel, which I'll now send. So our research agent is active again. It is working on two things now, and the second one is the Slack research. So it received our message from Slack. And with that, we initiated the next task. Meanwhile, our data-wise agent created

[5:01] charts for our report, so this is done. And now it wants to publish this report to here now. This provider we have for the website, and therefore requires also approval for action on Grokbot's computer, which is not my machine, but the machine that is running here in the cloud. So I approve that. Once, researchy also received information

[5:23] that the report is up. We get a summary, and we get the link. And what we can also do now with the computer users. Open the page in the browser on your computer. We see that it opens the page directly in Chrome and pastes the link. And we can actually now use this browser

[5:40] like our own one. So we have access to this computer. This is not our own machine again. This is the cloud computer that belongs to the bots. And in the meanwhile, we also got the result for our USA versus Germany. GDP development directly in Slack. So this was answered in a thread in the

[6:00] corresponding Slack channel. We also have an updated report now, where the data-wise charts are also implemented in the. Report that was created. So again, the agents spoke to each other to now complete this task. This whole process of researching, analyzing, and creating reports can now be packed into a so-called skill. Skills are one of the core ideas for AI

[6:25] agents. To give them instructions on a specific workflow. So in our case, whenever we research something, we want researchy to start the research, then datawise should analyze the data, and the report should be built by the reportify agent. This is a workflow that is reusable and that we don't have to rethink about every

[6:46] time we initiate a research. So with this instruction now, the skill is generated. This takes a few moments. And what we also saw is that I orchestrated the whole process by the research bot here. This is one option. Two other options are that you either create a boss bot, so to say, a bot that orchestrates all the other ones with the Task to oversee the team, set priorities, and delegate tasks to the three other

[7:16] bots. And an alternative would be that you create a group chat because with a new chat here you can add researchy. Datawise and Reportify and those now have a common group chat. So whenever you initiate a new task, in this case, a research data analysis and reporting task, those three bots will orchestrate this all their own. It's a group chat, as you can think about

[7:40] in WhatsApp or Telegram. So here I don't mention any specific bot. They just have to find that out on their own. And you can always have only one group chat with the same participants. So I would be able to create a second group chat with Researchy and Reportify, for example. But I cannot have all three bots in

[7:59] another group chat. Again, the same logic applies. One chat with specific participants is limited to one chat instance. And here in our group chat, things are starting now. The bots, the bots talk to each other. Data-wise, got started with the Claude cowork and ChatGPT work research. And under plugins, we can find the private

[8:21] skill that we created. So this skill is our report pipeline now. So whenever we start the research, this pipeline will be executed. And this is what the skill looks like. It's basically a Written description on what will be done in case of such a research is initiated. You can edit skills in here. You can also delete skills. Currently, you cannot create skills

[8:44] manually. You always have to create them with a bot. The co-work report is also here in its first version, and as you can see, the styling is basically the same that we had for the Ferrari report. So I will update the styling of the last website that was created to the one that we have on acadmind.com, our website for

[9:05] courses on AI and technology in general. The bot browser then checks our website and analyzes it, and then hopefully takes the colors, layout, and the general styling. For our latest output that was created, and of course you could then also save this styling as a default styling for your reports to not always have to repeat that

[9:27] flow. This is now the updated page with the Academind scheme and styling, and this is Grokbot in a nutshell. It is an agent with its own server. You can spin up as many agents as you want to. They have the same shared computer. They can do everything you can on the computer. They can even install tools on that

[9:48] computer. They can browse the web. You can orchestrate them either in groups or individually. You can refer to other agents with the at command that we saw here. You can install plugins. You can run routines to make sure your agents are started when you, for example, send a Slack message. All this, of course, also works from the

[10:08] mobile phone. This is now the mobile interface, so I can access all my chats as I had it on the desktop computer. I have all my bots in here, and I can always use those, as those bots are running in the cloud on their own computer. Those are always on, so whatever I do on the mobile phone also works on the desktop

[10:29] computer, and vice versa. So you have your always available chatbot army now. And Grokbot is therefore a very interesting tool. It comes with a lot of nice convenience features, especially this chat interface is. Very nice to use. The fact that we have one single chat where we work is a bit new, but it worked

[10:50] pretty well so far when I tried it. And of course, the idea of spinning up agents whenever you want to—you have their computer which is always running. So whenever any task, any idea comes to your mind, you can immediately initiate that and let that task complete by your agents. So definitely worth a look.

---

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