Most AI tools sit in a separate tab you have to remember to open. Buzz flips that. It’s a team chat app — built by Twitter co-founder Jack Dorsey — where AI agents aren’t bolted on as an afterthought. They’re full participants in your channels, right alongside your human teammates.
If you’ve ever wished you could @mention an AI the same way you’d ping a colleague, this is the closest thing to that yet.
What Buzz Actually Is
On the surface, Buzz looks like a standard workplace chat tool: channels, threads, direct messages. But the key difference is that you can invite AI agents — Claude Code, OpenAI’s Codex, or Goose (an open model runner, also from Dorsey’s camp) — into any channel as named participants.
Each agent gets a handle. You tag them. They respond. And because they’re all in the same environment, they can read each other’s messages too.
It’s also decentralized, open-source, and currently free.
Setting Up Your Agent Team
Once you install Buzz, you’ll create an identity key and then add whichever agents you want. You can assign different models to different agent personas — one running Claude, another on Codex, a third on Grok — and give each a distinct name so your channel doesn’t turn into a wall of identical bot responses.
From there, treat them like any team member:
- Drop them into a channel with specific humans
- Create agent-only channels for automated workflows
- Mix people and agents freely in the same thread
The Real Power: Agents Checking Each Other’s Work
Here’s where Buzz gets genuinely interesting. Because multiple agents share the same channel context, you can set up peer review loops without any custom scripting.
Say your team is drafting a product brief. You could:
- Ask Agent A (Claude) to write the first draft
- Tag Agent B (Codex) to review it for technical accuracy
- Tag Agent C (Grok) to punch up the positioning language
- Ask Agent A to revise based on the feedback
All of that happens in one thread. You can watch the agents respond to each other, catch where they agree, and intervene when something goes sideways. It’s a feedback loop that used to require either a lot of copy-pasting between tabs or custom API orchestration.
The agents will even start addressing each other directly — one acknowledging another’s critique, then elaborating on what it would change. It reads oddly like a real team conversation, which is either impressive or unsettling depending on your disposition.
Practical Ways to Use This Today
A few workflows where this setup earns its keep:
Code review across models. Push a function to a shared channel, tag two agents with different coding strengths, and get competing critiques. Disagreements between them often surface the most important edge cases.
Content iteration. Have one agent draft a landing page section, another critique the clarity, and a third tighten the SEO angle. One thread, no context switching.
Decision support. Describe a tradeoff your team is debating. Ask three agents to argue different positions. Use the output to structure your actual human discussion.
Async task delegation. Assign an agent a research task in a channel, go do other work, and come back to a threaded summary — the same way you’d delegate to a junior teammate over Slack.
What It’s Not (Yet)
Buzz isn’t going to replace Slack for large organizations tomorrow. Enterprise features like SSO, audit logs, and granular permissions aren’t the focus right now. The agent responses can also be slow when you’re running heavier models — expect to wait if you’re using something like Claude Opus on a complex task.
And the agents don’t yet have persistent memory across sessions the way a human teammate would. Each conversation is relatively fresh. That limits how much institutional context they carry.
Worth Trying If You’re Already Using AI Agents
If you’re regularly bouncing between Claude, Codex, or other tools for different parts of your work, Buzz gives you a single place to coordinate them — and bring your human collaborators into the same space. The peer-review loop alone is worth an afternoon of experimentation.
Start small: one channel, two agents, one repeatable task you do every week. See if the threaded back-and-forth saves you time versus your current workflow. If it does, you’ll find more uses quickly.
