October 6th, 2026
From Hermes to Grok Bot
A few months ago, I wrote a post that sat at the top of Hacker News for a day, describing my Hermes agent with Block Buzz setup. I spent a month setting it up, fine-tuning it, and making it a tiny bit useful. The engineer in me felt so satisfied and others seemed to share that excitement. Coincidentally, within a day of my post, xAI announced Grok Bot. At first I brushed it off as a wussy option for normies. Then I tried it and within a few days I deleted my entire agent setup and switched. Why?
The Pain of Hermes
In short, I spent 90% of the time tinkering and fine-tuning Hermes+Buzz and 10% getting actual use out of it. I was managing the Hermes agents, managing the VPS they run on, managing this new thing called Buzz to communicate with the agents, managing the MCP and connection statuses of agents, managing the cron jobs to ensure they fired properly, trying to figure out how to give my agents a proper browser, and on and on (trust me, the on and on really is on and on, but I said "in short" so I'm sticking to it.)
All that stuff is solvable, but I felt like I was justifying my 30 years of engineering experience rather than focusing on what it could actually do for me. So I installed Grok Bot, asked it to remind me in a week to delete my VPS, and the coup d'état had begun.
The Joy of Grok Bot
Grok Bot took 5 minutes to install, and a few more minutes to set up all the core connections, some of which were in the marketplace, and some of which used its shared computer to connect and share access across bots. A few more minutes and it was on my phone, storing reminders, recurring reports, and doing research. Everything just... worked. No hassle, no fuss, very little tinkering. I was both joyful, and at the same time, that feeling of "getting replaced" kept popping up in my head ("but if normies can do this, what use am I?"). That's why over the next few months I'm going to spend a fair amount of energy and tokens expanding the usefulness of Grok Bot.
What it's not
So far I'm not thrilled about Grok Bot's engineering capabilities. I find myself opening Codex and Claude Code for most of that still. Grok Bot certainly CAN engineer and it farms work out to Cursor Cloud Agents and you can even try and pin the model for those so I'll pin the frontier models, but I still find it makes more simple mistakes. That really shows how important the harness is. I'm even kicking around the idea of having Grok Bot farm work out to Codex running in the cloud.... but that requires managing the wiring and that's what I don't want to do.
The Next UX
Technology innovation always happens in this flock-of-birds mentality where everyone flies in one direction, then a bird starts darting a different direction, and the flock follows, eventually reforming back together. You can see that happen with the UX of AI right now from API, to chat, to IDE+chat, to CLI, to desktop app, back to CLI, to local agents, to cloud agents, to Grok Bot. Now we're seeing the wave of Meta's Muse, OpenAI Dots, Instinct, and I assume the Earendil folks have something?
In any case I'm excited to see where things go, and Grok Bot definitely feels at least directionally correct.
100 Days of Grok Bot
Here's my idea: over the next 100 days, I plan on exploring and posting use cases for Grok Bot, one use case per day. Some will be small, and some large, and I definitely don't have 100 now so there's some faith that I'll find them. Feel free to post your own using hashtag #100DaysOfGrokBot. Let's see how it goes!
Here's a snapshot of some of those use cases that I'll expand on a bit:
- Website modifications
- email list management
- off-computer backups
- daily social search
- daily events search
- video creation
- daily reminder system
- and 93 more... hopefully.
You can follow my X account for the 100 days of Grok Bot. I'll also create a new Blog post and edit it every day with the new use case.