Grok Bot Team Management Checklist: Hire Fewer, Let the Bot Write Its Own Skills

Grok Bot is xAI's bot system. Users can add bots to their own team, assign work with @ mentions, and add bots made by others from templates. X user @poteto (lauren) posted a thread about how she builds high-quality bots and manages her team. At the end of the post she left an entry point: have @bot read this post and help improve your own team. A bunch of people in the replies did exactly that.

First, the tools she uses. There is a bot on the team called dr eggbot, positioned as a designer, used to create other bots. The template instructions are quite specific: ask a few preference questions first, then start building. Coding bots get a standard called poteto-mode: one responsibility, no fudging, verifiable. Non-coding bots get the same treatment: one responsibility, one voice, clearly state what not to do, and don't attach tools they won't use. The tone is casual, a bit mad-scientist, short sentences, all lowercase, inclined to just start acting once the task is clear, and by default does not output shareable templates.

Then there are the team management points.

You don't need many bots. A few good ones are more useful and easier to manage than a large squad. This runs counter to many people's intuition. Cobalt in the replies later added the same judgment: the more bots there are, the more coordination problems there are, not more output.

Have it write its approach as a skill. When you teach it to do something, also have it create a skill for itself. In long conversations, it relies on this to stay consistent. If you want to keep things more independent, have it create a private GitHub repo and store all the skills there.

Give it full tool permissions. Connect connectors, use the vault, and authorize all the tools you use for work. You can also have it register its own email and open its own accounts, separate from your identity.

Look for the copy-paste work. Think about which steps in your workflow require copying information back to the bot, and just ask it whether it can do that step for you.

Train it the way you'd train an intern. Imagine you hired an intern: how would you introduce the work, and how would you get them to do it right? First introduce yourself and what you do to the bot, then ask how it can help you automate.

Use voice. The original author's favorite item. Voice mode or transcription: dump a pile of information all at once. At the end, add: repeat that back in your own words. Confirm it actually understood, then move on.

There were a few practical pieces of feedback in the replies.

Bardya said voice mode is very useful, and also uses dr eggbot as a weekly auditor for his entire bot roster. In the screenshot he posted, the voice call with dr eggbot lasted 47 minutes and 51 seconds.

Vishal raised a question: when speaking a mix of Chinese and English, the bot gets stuck. This was forwarded by the original author to a colleague.

Cobalt's addition deserves separate attention. He said the intern analogy is very accurate, but don't give every bot every permission just because you can. Each bot should have one clear responsibility, limited permissions, and a way to prove it actually finished the work. Skills solve consistency; verification solves trust. His conclusion is that the goal is not the largest bot team. The smallest team that can reliably get the work done is enough. The original author replied that these points were good.

Kent's approach is to have the chief bot on the team suggest improvements to itself; he said the suggestions were quite solid.

Someone else asked about a detail: when dr eggbot is updated in the future, will the copy already added to their team change along with it? Opinions in the community differ, and he is waiting for the original author to answer.

The step most easily skipped in this checklist is looking for copy-paste work. People are willing to spend time tuning prompts and configuring permissions, but rarely stop to list what they copy and paste every day. And that is exactly where bots should take over. Another easily overlooked point is 'clearly state what not to do.' dr eggbot's instruction page writes it alongside responsibilities, and its effect is no less important than responsibilities.

Original post: https://x.com/poteto/status/2116576229319233596
dr eggbot template: https://x.ai/bot/_jOdbfkB16zxu7MRcmReE