Using Grok Bot: Group Chats as Project Rooms

You may have seen people treat their agent as one chat window that does everything. Here’s how I’m running Grok Bot with group chats instead. And it's also why I haven't completely adopted ChatGPT's Dot.

Building the rooms

I don’t run one mega agent.

I run a room per project.

Right now that looks like:

• Specialist Desk

• • Project 1
• • Project 2
• • Project 3

How the stand-up works

Each room gets the bots that own the work.

Navi routes. PM tracks. Design and build stay in the thread with the people who do that craft.

Each agent still has its own computer.

That matters. The bot can open files, run tools, and keep state without sharing one desktop.

The group chat is the stand up, not the brain.

Short updates. Clear lanes. Blockers in the open.

Rules I force on the system:

AUTO for research and drafts.

NEEDS ME for anything that posts, sends, or spends.

If a bot can hit publish without me, I don’t have a workflow.

I have a liability.

Watching it cook

What’s been working this week:

Approvals (AUTO vs NEEDS ME) as the control layer.

Side projects moving in rooms instead of DMs.

Notion as the source of truth when a room hits the seat limit.

Grok Bot’s 6-bot room cap is real.

When a specialist gets bumped, the project doesn’t die.

I have a Command Center (overview) keeps the record. The room stays for the live work.

## A few meta-points

1. One agent per job beats one agent that pretends to be a company.

2. Group chat without ownership is Slack theater with better typing. (Tokens say goodbye here)

3. The receipt isn’t the model. It’s the control surface and the room roster.

The unlock wasn’t a smarter model.

It was seating.

What I have now is a durable pattern I can reuse:

Room → roster → stand-up → governor rules → Notion when the room overflows.

Same shape for a new product, a build, or a deal search.

Swap the seats. Keep the rules.

If you’re still stuffing everything into one agent chat, try a room.

Chatgpt's Dots needs to expand it's main surface.

Seat the specialists agent. Keep the computers separate. Make publish a NEEDS ME moment.

Who else is running multi-agent rooms like this?