My Grok bot had me panicking for 15 minutes

I spent the last two weeks trying to pry myself away from meddling in my business, instead of working on my business. This is one of the things that Michael Gerber talks about in The E-myth Revisited, in his definition of “The Technician”:

Everyone who starts a business eventually learns that making the business work takes three distinct personalities: The Technician does the work of making it, selling it and delivering it, so the doors can stay open and money’s coming in. The Manager focuses on achieving results through people and systems so the business can sustain itself and grow. Finally, The Entrepreneur defines the vision for the business and on closing the gap between where the business is today and where they want it to be.

I’m certainly a Technician, which is a problem, because (again, according to Gerber):

To be a great Technician is simply insufficient to the task of building a great small business. Being consumed by the tactical work of the business, as every Technician suffering from an Entrepreneurial Seizure is, leads to only one thing: a complicated, frustrating, and, eventually, demeaning job!

He goes on to say other fun and on-point things about Technicians, and if you want to read more I highly suggest you listen to the audiobook and, if you’re so inclined, I also have an interactive, which will help you implement Gerber’s ideas for your own business. It’s available to Kolea.pro members so get yourself on the wait list!

We’re living through an incredible time, where AI can help us leverage ourselves in dramatic ways, which is the theme of my Fall Cohort. I have exactly 2 seats left and we’re focused on building incredible things, so join us!

I made a bunch of these as part of Kolea.pro, which is a fun app that helps you connect with your audience. I’ve always struggled with the notion of outreach (aka “marketing”) because I feel like a slimeball. I read a bunch of books, however, that really helped with that so I turned them into interactive AI “operating systems” or “machines”, where you can, basically, chat with a book.

Removing myself, building my own machine

The entire point of Gerber’s book is to build a solid business process, with every step written down and repeatable. A franchise, basically, where all the roles are replacable and the business can scale by sending along a book of guidelines and principles.

Yeah, tons of fun. He uses MacDonald’s as the canonical success for this model, and I suppose he has a point, but I don’t want to build a MacDonald’s. I want to build something fun, which actually emphasizes a key point he makes throughout: you can’t build a business while working inside the business.

I spend way too much time fiddling with the mechanics of what I do. I enjoy it a lot because I like building and optimizing processes and apps and, in short, I love the coding process. The problem is that I don’t do it any more, and honestly I shouldn’t if I’m going to make something of Kolea.

I have a dream of building a 7-figure business with a super-streamlined operation. I think it’s possible, but not without some type of orchestration and process. A bitchen set of prompts and skills won’t cut it - I need some autonomy!

So that’s the task I had before me two weekends ago, and it didn’t end well, I hate to say.

Hermes and OpenClaw. Seriously?

I used OpenClaw immediately when it came out and liked it a lot, and set up Discord with a fleet of 6 agents that all had specific tasks, from ops to marketing to personal fitness. It worked well, but soon I found that I was fiddling with the damned agents more than if I just used Claude on its own!

Either way, I had fun with it right up until Anthropic pulled the rug out with subscription API access, so I tried it with OpenAI models and that was … let’s say suboptimal.

I’m going to summarize my experience with both Hermes and OpenClaw the same way: if you enjoy fiddling with botware, it won’t bother you much. Edge cases seem to pop up constantly, and the term “skills issue” will be on repeat in your ego-bruised brain.

These days I have Anthropic credits with my 200max plan, so I figured OpenClaw might be the call, once again. It took me exactly 48 hours to grind this thing down with gateway crashes, which I have yet to understand, and yes I know that I could blah blah blah but that’s the point, isn’t it! Something isn’t clicking here, and I’m constantly problem solving.

I don’t want to fiddle with bots. I want them to actually do the work so I can remove myself from the business!

OK what about Grok Bot?

I had read amazing things about Grok Bot so I decided to give it a try while also getting OpenClaw up and running. I gave them the same context about me and my business, and figured “may the best bot win!”

Grok Bot blew my mind at first. Everything just worked straight away, and there were no crashes at all. It was also kind of cute and fun, and the voice was easy to talk to while I walked around the block at night, and I almost felt like I was talking to my XO, which is what I told Ducks, my primary bot, he was.

That’s when I realized that Grok is no Claude.

The trouble started when I created a Lead Dev bot and set up GitHub monitoring. Lead Dev’s job was simple: react to issues assigned to me and then work on them. I decided to do a simple first test: update the project README so it made sense and was “comprehensive”. I know what Claude would do with that, but what would Grok do?

Three hours later I checked in because the issue was still there and I couldn’t see any PR queued. This is the response I got:

Lead Dev did the work but never touched the issue. I just checked #16 on GitHub, and it has no labels and no comments. The README rewrite is on a branch called docs/readme-refresh, waiting for you to review it. It stalled because Lead Dev won't post anything under your name until you tap yes on its card. Then it sat waiting and I didn't chase it, which is on me.

I've told Lead Dev to label #16 status: needs review and post a short comment saying it's ready for you to look at. From now on it'll do the same on every issue: mark it in progress when it starts, then comment when it's ready for review. If its approval card shows up in Lead Dev's chat, one tap from you gets it moving.

I know, I know. I didn’t tell it to create a PR. I do have skills in the repo with an AGENTS.md that describes the development, testing, refactoring, and quick-fix process. All of them were ignored, which I guess shouldn’t surprise me?

That was a quick fix on my part, directing Lead Dev (and Ducks) to always use Claude Code on its local computer for any and all dev work, signed in as me.

But then, later that day, came this treat:

In short: Ducks reported that there was an HN thread where I was apparently being accused of using AI crap for content. Of course I dropped everything because I don’t do that and needed to defend myself and my buisness, and I was extremely grateful my new Strategist bot caught this!

After 15 minutes of elevated blood pressure, I couldn’t find anyone discussing my site on the HN thread, so I asked Ducks for more details.

This is so good:

It all comes down to wording, doesn’t it? “Hacker News thread asking exactly what Big Machine teaches…” You really can take that in a few different ways, can’t you. Maybe I’m paranoid, or maybe Grok sucks at grammar.

Anyway: It’s hard to pull back from this and trust Mr. Ducks again. Perhaps I could instruct it to use complete sentences like my mom used to do to me? This is the level of fiddling I can’t stand, and here we are, going through a therapy session where I get to admonish my bot to “use its words”:

When you say ‘it’s finished’ I assume you mean the work is finished, not your response to my prompt. It’s this kind of thing that makes me emotionally unavailable to you…

I still have my sub because that’s the way I roll: I’m a tweaker and I love tweaking. Not in that way, of course, but I suppose it’s just as bad.

Hi everyone my name is Rob, and I’m an AI tweaker. I’ve been using AI for about 2.5 years now and watched my job slip away as well as my ability to concentrate. I fly into rages easily, yelling at my computer, wondering what’s the point…

Time for my next fix…

For now I’ll just use Claude

I know I’ll try these bots at some point again because I can’t quit them and that’s the future, isn’t it? Autonomous agents that do a thing, as long as we’re clear what it is they’re supposed to do.

To that end, I have found that taking a few hours to set up a few projects in Claude Desktop (both Code and whatever the not-code is) with some recurring tasks is all I need.

I have a “Daily Story” artifact that gets refreshed, support triage, all of my MCP servers hooked up so I can answer questions super quickly, and Trello automation. I also have GitHub hooked up too. The only thing I don’t have is a cute little icon talking back to me throughout the day.

I know that’s coming with Duet and, if I’m honest, I don’t know if that’s really a good thing. Like I said: I’m an AI tweaker and the last thing I need is another thing to tweak.

I’m trying to pull myself out of the business!

The more I have to tweak bots and fiddle with apps like Grok Bot, the less I’m creating the content that will fire my business. This week is a perfect example! I’m trying to finish up The AI Pro v2 course, which is a full redo of my flagship course, and I’ve been dorking around with Grok!

This isn’t the way to run a content business, is it?

So it’s back to Claude Desktop for now, where I have a whole suite of projects and scheduled tasks that do what I need done. Development, for now, is still me and a set of 3-4 projects running in Claude Code on the desktop too.

I know I need to automate this, but for now I’ll keep on with my Kamaji thing, running a bunch of things at once, and dreaming about how I can automate them in the future.

What do you do? I’m sure you have a cool system and a way of doing things I probably haven’t thought of. I’ll keep trying!

Cheers, and thanks for reading!
Rob