# I Replaced My AI Team with Grok Bot and I'm Not Going Back

Channel: ICOR with Tom | AI Productivity
Video: https://www.youtube.com/watch?v=_UyAz027uIw
Duration: 29 min
Language: English
Words: 4691
Transcript page: https://viewrankai.com/tools/youtube-transcript/_UyAz027uIw

---

[0:00] My entire AI team now runs inside Grok Bot, and the lead agent inside Grok Bot I call Larry The one I talk to who hands the work to the right specialist. And if you know this channel, you know that's a big deal because not long ago I said this. And that's why we will stick to this method until we see that there's really an AI coming out that does it as good as our local folder approach

[0:26] well, I think that moment came, and I know exactly when it happened. I was sitting in my car. I tell you that story in a few minutes because it's the reason I won't build this myself again So let's dive into this. If you're new here, welcome. Quick catch up for you. First, who I am. For eight years, I worked in big corporate, back-to-back meetings

[0:51] and everything that comes with it. In August 2021, I left to run myICOR full time. that's when I started coaching business owners, solopreneurs, and freelancers to become more productive at their job, passing on all the productivity insights that I gained climbing the ladder in corporate. So I'm not here as a typical YouTuber. I'm a busy professional who has to get more done in less time, just like you. My co-founder, Paco Cantero, runs several businesses, and the two of us came to the same conclusions and used tools like Grok Bot to optimize

[1:29] our own work and our own lives. Over many months, we built something we called myPKA, my personal knowledge assistance. and later, we named it My Personal Knowledge Agency. The idea was simple. Treat AI like a human team. One lead that I called Larry back then. He doesn't do the work. He delegates. And a team of specialists, each with one clear job.

[1:55] today the big companies are doing exactly this. They call it orchestration. And here's the thing, that's not a word I picked up from them. I called Larry my team orchestrator from day one. Back in spring 2026, we were already talking about exactly what the big companies are building now in plain folders on my Mac.

[2:17] It worked. It worked so well that I told you I'd never move it into a hosted bot, and I gave you my reasons. Three worries: one, talking to one bot makes no sense when you run a whole team. Two, in the cloud, I lose ownership of what I create. And three, transparency. I want to see what my team knows, which rules it follows,

[2:44] and why it does what it does. All three were real, and I answer each one on screen in this video. Stay with me for number two, because that answer surprised me, too. But first, you need to see what I'm replacing. because this was not a small setup. Everything lived in one folder, and many of you watching this probably

[3:07] have a similar folder already, too. That's why this video might be really interesting for you too. Inside this folder, there's an AI team folder, and you see here 59 agents. Each one has its own folder. Each one has its own instruction file with its role, and each one has its own journal where it writes down what it learned working with me. Next to it, a team knowledge folder, standard operating procedures, work

[3:35] streams, guidelines, how we name files, how we write, how we publish. And this wasn't only my business. Larry and the team managed my whole life. Everything was connected in that one folder. All my health data, all my finances, every scanned document and contract, everything inside this local folder. Claude hooked into this folder and worked by these instructions.

[4:03] That became my orchestration layer. And over time, Claude took more and more of it on itself by adding sub-agents, and it worked perfectly. So when I asked Larry for something, he read the right SOP, picked the right specialist, briefed them, and put it all together. All plain text files, no code, nothing special going on.

[4:25] And to me, this was the dream. Fully local, fully mine. I could open any file and see why my team did what it did. And it worked in Claude, in Gemini, in Codex. Same folder, any model. That's how I solved transparency back then, through Larry. Every lesson went into a journal. Every rule went into an SOP.

[4:49] Every repeating job became a work stream. The thing is, at some point, my team started to overthink. I gave my full team a simple job, one overview page for a course that I want to create. Eight hours later, millions of tokens and a broken result. That is something where I then started another one, a plain session in an empty

[5:14] folder, 15 minutes, and better results. That was the Opus 5.5 pivotal moment for me. So my old folder became exactly like a corporate meeting where everybody has to sign off and nothing moves. Well, now in corporate, there are teams who build orchestration at big company scale. They have whole departments to keep all those rules in line.

[5:38] That's real effort. Far more effort than you and I need. I'm one person with a business to run. I don't need to be the one who maintains the bureaucracy of my own AI team. So here's what changed. In myPKA, I had SOPs and CRON jobs and other things like that. In Grok Bot, that's now skills and routines per bot.

[6:02] Same idea, but far more sophisticated and reliable. Look at the numbers. My AI team folder alone had over 37,000 files. 37,361 to be exact. Yes, about 5,700 were just installed code packages, but the rest was my team. Fifty-nine agents, each with its own agent.md, its instruction file, plus three retired ones. Hundred and forty-nine SOPs, fifty-seven work streams, hundred and seven guideline files, twenty skills, one thousand four hundred and

[6:40] eight task files, and two thousand two hundred and sixty session logs. And that just shows how much I was using this folder and the team actually, and that's after I had already cleaned up things. It used to be even more. Today, about two dozen bots, twenty-one shared skills, and three routines. Three for now.

[7:03] In the old folder, it kept growing. Lesson after lesson, rule after rule, until the rules started to fight each other. One SOP said that, one guideline said that. So at that point, the solution was force it through code. Scripts, not just plain text files. And that's what we did. More and more, it became a mix of both.

[7:25] Text files for the thinking and proper code for the rules that had to hold. That compensated for the issues a bit. For a while, until people started building their own tools around all of it. And that's all fine, but Grok Bot just simplified all of this. That's the point. I have to acknowledge this.

[7:46] Time is moving so fast right now. And here's the thing, a markdown file only works if the AI finds it and reads it. Maybe it does, maybe it doesn't. In Grok Bot, each bot's instructions is loaded every time. The skills are listed, so the right one gets picked for the task. And some rules aren't just written down anymore, they are enforced.

[8:10] My database connections read only on the connection itself. Risky things like sending or deleting get checked before they run, and some need my click. I had to manage all this manually on myself in my folder. So in my experience, my bots stick to the rules much better than my folder ever did. That's not a lab test.

[8:32] It's what I see every day. And now I want to talk to you if you're the one started with the local folder approach back in April, If it's working for you, keep using it. Don't break it immediately. Seriously. You will do more harm than it will help you if you don't know why you're moving on to something else. I'm only sharing what happened since then, and you can decide if

[8:58] you want to move on or if you're happy with what you have right now. But we have to acknowledge a lot has happened since April 2026, and that's the crazy times we are living in. It's insane. There's not one month going over where we see new AI releases, and this is where we need to stay on top of it. And that's where the backbone, the ICOR methodology, the tool-agnostic approach to all of these things make us resilient to these changes

[9:26] and make it easier to understand. And this ICOR methodology never changed. It's the same since four years ago when we launched it in myICOR But let's go back to April. Let's go back when the first time we revealed the myPKA setup and the orchestration idea. Because back then, openClaw was the autonomous agent everybody talked about.

[9:48] Then Hermes and others got big. But at the start, they were all one agent, one assistant in your chat. And that's why I went with Claude and sub-agents. And to me, that worked perfectly with orchestration. The main session handed the work to a specialist. The specialist loaded its own skills and its own SOPs at the start, and

[10:10] then it stayed focused on one job. At that time, that was totally efficient. But AI got better, fast. And for me, the pivotal moment was the release of Opus five point five. I realized I could stop using my AI team for a lot of things I just prompted Opus 5.5 directly out of the box, build me this application, update this part of

[10:35] my membership, and it did it perfectly. No SOPs required, not even any skills that I downloaded there. All the guardrails I had built file by file, the model now handles them itself. And then Grok Bot. Here's the honest part. In Grok Bot, I don't choose which model runs or how hard it thinks. That's not my call.

[10:56] But from my first experience, the output is spot on. and even that I worried, especially about this lack of control, I have to admit, the output is spot on. So this means the routing, which bot uses which model when, seems to be handled very well. And Elon Musk actually posted about this himself. He wrote, "Going forward, SpaceX will use the best back-end model for any given task, including Claude Opus 5.5, and Midjourney, and even

[11:28] Suno, and other leading APIs." So things got more complex on my side. At the same time, AI got much more efficient and much better at orchestration out of the box. The lead, the specialist, the handover, that's built in now. So the local folder structure, the way we had it back then, to me, is not needed anymore.

[11:52] And I have to say that openly. That's part of what this video is about. The thinking, the concepts behind this stays. The folder were just a way we could do it back then. So I moved the whole team into Grok Bot. some agents I merged because they always worked together anyway. Some new specialists came in.

[12:14] Today, I have about two dozen bots in Grok Bot. Each one has one job. By the way, in Grok Bot, your first bot, the orchestrator, is simply called Grok Bot. But you can rename every bot however you like and even change its picture, so it really becomes your own setup. minus, as you can guess, Larry, of course. Honestly, I really like the animated avatars now, so I don't feel the need to

[12:39] change the looks at all or bring back the animal images that I used in the past. For the other bots, I kept the names descriptive: finance, health, so it's clear what each one is about. Will I give them other names again one day? I'm not sure yet. and how about the move itself? I didn't copy fifty-nine agent folders over by hand.

[13:03] I went through it with my new Larry and Grok Bot one role at a time. Who stays, who merges, with whom, what goes into a bot's instructions, and what becomes a skill that every bot can use. And here's the part I still can't quite believe. I did all of that while I was driving in my car on my way to the

[13:25] supermarket, just talking to Grok Bot. Who stays, who merges, who goes where, and when I come back home, everything was already set up perfectly. The handover was seamless. I sat down at my computer and just kept working, And this was a revelation for me. And you see here, it's the same idea as my folder.

[13:46] Larry is still the lead. Writer, course designer, YouTube helper, researcher. Each has its own instruction, its own memory, its own journal, and I can talk to each of them one-on-one. And I have to admit, that's new for me. In my folder setup, I had one rule: I only talk to Larry, never to the specialists directly. I trusted Larry to orchestrate better than I would if I

[14:13] jumped in and interrupted them. With Grok Bot, that changed. Now it makes much more sense to go straight to the bot for the topic. My finance bot, only about taxes and accounting. And when a job needs several of them, like preparing this video, for example, I bring them together in a group chat. so I'm not only talking to the orchestrator anymore.

[14:34] But Larry is still my first contact when I have no idea where to start. He's still there to work through open requests, to catch anything the other bots missed, and to get me back on track when I got lost. This is myPKA. Skills, memory, journals, one-on-one chats, all the same pieces. The difference? I didn't build any of it.

[14:55] Grok Bot does it on the fly. Now let's answer the worries, starting with number one. I thought Grok Bot means one assistant, one chat, one person. Well, it's not. And honestly, that was my confusion at the start. I knew Grok Bot had different agents, But I was used to Larry being my one and only point of contact. So the idea of talking to bots one by one, or to an orchestrator, or to a whole

[15:25] team in a group chat actually kept me away from trying it in the first place. but the thing is, it turned out to be much more intuitive than I initially thought, and I'm not looking back. another big difference is in my last setup, Larry was talking to the sub-agents, and the sub-agents were only reporting back to Larry.

[15:45] Here, it's different. Inside Grok Bot, the bots can message each other directly too. That's exactly what Larry did to the sub-agents, but not between the different sub-agents And in Grok Bot, I didn't build it. I don't maintain it. It's just there. And when I need just one of them, I open my one-on-one chat, like with a colleague.

[16:08] The writer knows my voice. The YouTube helper knows my channel. I don't explain it every time because they remember. So far, I run eight of these group chats. Might be more in the future. One for video prep, one for editing, one for a course. They replaced my work stream files in the past. So worry number one, done.

[16:31] Now before worry number two, let me show you what I actually get. Because some of this I tried to build myself and some of it I never finished. One. One computer in the cloud. All my bots share it and each bot gets its own screen. So it's not a computer per bot. One computer, a screen for each bot.

[16:54] the files are shared like it was in this one local folder before. One bot picks up what another one made. For this video, the researcher saved a fact file. The writer opened that same file and wrote from it. Nobody had to send anything around. Two, it's always on. I close my laptop, and my bots keep working. In my folder setup, when the Mac slept, Larry slept, and

[17:20] my routines run on their own. My support ticket helper sends me a briefing every weekday morning at 8:00. Number three, every device, Mac, Windows, Linux, iPhone, iPad, Android, same bots, same chats. And number four, files. When a bot makes something, the file shows up right in the chat. I preview it, I save it if I want.

[17:45] And it goes further. It can push everything to other tools Or anything I need to review or videos, I can push to Dropbox or G Drive and directly access it there. Five, my tools. email, calendar, my self-built app AYNIC, Dropbox, ClickUp, Todoist, and so on. I connect them once for my account and my bots can use them.

[18:08] And for anything without a connection, they use the browser. Number six, memory and journals. Each bot remembers its own facts and keeps its own journal. And some facts reach every bot, like files go to Dropbox. I said it once. Done. Point number seven, templates. A bot set up its instruction, its memories, its skills. I can package that as a template and share it with others, without my

[18:37] password and without my chat history. And eight, voice. I can dictate, I can have a live voice chat, and my bot send me voice memos back I can have real conversations while I'm reviewing something. It's waiting for me until I finished reading. It's not interrupting me. I can have a constant conversation with a bot to keep working,

[19:00] That's really a big thing too. That's months of my own building partly, and some of it I never got to Now, a list of features is one thing. What surprised me is what it did to the way I work. Three things. First, I stopped being the copy and paste person. Before I moved text from one tool to the next all day. Now the file just arrives in the chat, in my Dropbox, or in

[19:28] my tool, whatever it might be. Second, fewer bots, clearer jobs. I went from about sixty agents to about two dozen bots, and every bot is easier to explain. If I can't say in one sentence what a bot does, it doesn't need to exist. Third, I check results, not processes. Before I watched every step because I didn't trust the chain.

[19:51] Now the risky steps come to me as a question. Send this, delete that. I decide, and the rest just runs. But voice is where it really got me. Which brings me to the car. So let me tell you about that moment. I was in my car, and I had a live voice chat with my bot, just talking,

[20:14] planning this exact video Then I arrived. I sat down on my MacBook, I opened the app, and the conversation was just there, same chat. I kept typing where I had stopped talking. No copy and paste. No, let me sum up what we discussed. It just continued, and that's when it clicked. Because that's how I actually work, and might be you too.

[20:36] Ideas come when I walk, when I drive. Before they ended up in a voice memo, I maybe not listened to again, or I handed it over to my team to process in some way or form. Now they end up in a conversation that's still there when I sit down, but it's already done because it was processing while I was talking to the agent.

[20:57] And yes, I could build this with Claude Code. I've built a lot with Claude Code. I even built my own chat app around my team, but I said it myself back then. You get to eighty or ninety percent very fast, and then the last twenty percent is where the real work comes in. A voice call that hands over the typing on another device in the

[21:18] same conversation without a break, that is the last twenty percent. It takes a long time to build, and I won't build it again. So is Grok Bot perfect? No. It has its limit like any other tool, even the things that I built for myself. But to me, today, it's the best in class. And I say that as somebody who built orchestration himself

[21:40] back when nobody called it that. So now, worry number two, ownership. This was the big reason I said no. I thought hosted bots decide what they store for you. So before I moved anything, I checked. And you see here, it’s files. It’s folders. My bot’s work sits as files on their computer. And if I want everything, I just ask.

[22:07] My bot can zip it all and send it to me. Now let me be precise, because you’ll ask in the comments. There’s no single export button. That’s not how it works. But everything is files, and I can get those files whenever I want. And a bot setup I can package as a template and take with me.

[22:27] And one more thing, because it matters. My bots don’t decide what to keep for me. I decide. If I want something kept, it’s a file, or I tell the bot to remember it. If I want it gone, I say so. So in the end, am I locked in? Not more than I was with my folder.

[22:44] The folder just lives on a different computer now. And now worry number three, transparency. Same answer. I can open a bot’s instruction. I can see its skills. I can read its journal, and I can see every file it made. Plus, fewer rules means I actually understand them again. So worry two and three, done.

[23:09] Now, very important. One thing I really want to say, this is not stop building your own tools. Not at all. I still use my own app every day. It's called AYNIC, as you might know. My knowledge, my own knowledge app that I even use for task management and so on. I build it myself with Claude Code following our own course, and Grok Bot plugs into it through an API, an MCP connection, the

[23:37] way we teach in the course. So I just tell my bot what to add and where it belongs, and it creates the entry in AYNIC and puts it in the right place. I talk to Grok Bot, and my own app gets organized. And this matters to me. My knowledge, my people, my notes, they live in my own

[23:56] app in a structure I designed. The bots work for that structure, not the other way around. And if one day I switch bots again, my app, AYNIC, stays exactly where it is. And that's really the point. It doesn't matter which tools you use, your own app, your own app plus a few others, or only tools out of the box, like Todoist and Notion, or

[24:20] Obsidian and Todoist, whatever you like. Grok Bot is just a layer on top. I orchestrate my work from there, and the content still lives inside those tools. And that's important. Yes, I talked a lot about transparency and about seeing everything in Grok Bot, but you can still store anything directly in the tool it belongs to.

[24:43] That's exactly what I do with AYNIC. I keep my own database up to date with what I want to keep in place. Invoices, transactions, my journal entries, all of that still ends up in my own applications, and that's the key. What Grok Bot actually does is take away the friction, capturing, organizing, even taking action. I just say, "Create a new calendar entry," or, "Look up my Gmail." And when I'm on the road somewhere, I talk through my open items, and Grok Bot and the team

[25:16] handle it on the fly in near real time. To me, that's the best of both worlds. My own tool for my own knowledge and a bot team I don't have to maintain. And one more thing, just so it's clear, I'm not affiliated with SpaceX, and this video is not sponsored by them in any way or form.

[25:35] Nobody paid me for that. It's just progress in this crazy journey that I want to share with you. This is my personal journey, and I'm simply sharing it with you But I know some of you are different. You want full control over your data, over the model, or you have a Claude subscription and you want to keep using it.

[25:57] That's perfectly fine. And for you, we have a course in the myICOR membership. Build your own agent harness, and it's live already. There we will build exactly the kind of system we had in an optimized way That will feel similar to the Grok Bot setup too, but with a lot more control. Your own lead agent, your own specialist on the Claude Code engine

[26:21] with your data where you want it. It's a great way to really understand how all of this works under the hood in Grok Bot. Even if you end up using a hosted bot like me, you'll know what's going on inside. Me, I'll stick with Grok Bot for now because it's just easier to use,

[26:40] And I have more important things to do than nurturing my team. They need to nurture me. And if you want to start or you're already using Grok Bot, we also have a new course for that as well, and it's already live inside the membership, Your AI Team with Grok Bot. There I show you. There you see step-by-step how to set up your own team of bots just like mine, your lead bot, your specialists, your group chats, without building

[27:08] any of the machinery yourself. And also best practices using Grok Bot, how everything is connected, what you need to take care of and so on. And one quick note, because you might have heard about Claude Duet too, and what the heck, Tom? You're now talking about a complete different platform and no longer about Claude.

[27:25] Well, it's a waitlist preview and only in the US for now. So here in Germany, it's not an option for me. But you can be sure we will review it and compare it to Grok Bot the moment I get my hands on it. So let me bring it together. The idea never changed. The idea of myPKA, the orchestration never changed.

[27:47] Treat AI like a team. A lead, specialists, clear jobs, their own knowledge. That's ICOR, even if you're working with human teams, and that stays. What changed is who carries the machinery. For a long time, that was me, or you when you did the same. Now it's not. If you followed my folder videos, nothing you learned is wasted.

[28:11] You use the same thinking. You just don't have to build the house first. And if you're starting today, start small. One lead, two or three specialists, give each one exactly one job. And if you want to migrate, do it like me. Just point them to your old folder or any other setup that you have

[28:29] with your AI team and ask how it can be migrated into Grok Bot. Also important to write down what each of those bot should never do. That last part matters more than you think. Then let it grow only when you feel the need, not before. That's the lesson my 149 SOPs taught me. so now I want to know from you, would you rather build your own harness

[28:57] or hand over the upkeep like I did? Let me know in the comments below. Let's chat about it and subscribe to the channel so I can catch you up in 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