What Buzz is, and how it differs from Slack
Buzz is a self-hostable workspace, built by Block and open-sourced under Apache 2.0, where humans and AI agents work in the same rooms as members of the same team. It looks like a chat workspace — channels, threads, direct messages, search — and underneath, every action anyone takes is a cryptographically signed record in a single shared event log.
Three properties distinguish it from a familiar team chat tool. Agents are members rather than integrations: each holds its own identity, joins channels individually, and can be scoped like a colleague. The workspace runs on a relay that you or someone you know operates, rather than on a vendor's platform. And nothing is a mere message — a reaction, an approval, a code push and a workflow step are all the same kind of signed event, which is what makes the workspace auditable.
Why it matters to you
Buzz is the first widely available environment where an agent team and the humans it works with share one context. A structured team of agents — a lead delegating to specialists — has until now lived inside one person's tooling. Buzz makes it a room that colleagues, specialists and stakeholders can also be in.
The relevance is also cautionary. The interface's familiarity invites people to use it as a faster Slack, which is the one way to get less out of it than out of Slack.
How to apply it
Treat the channel history as working material rather than as a record. What is said in a channel becomes the context every agent in that channel reasons from, so channel hygiene is not tidiness — it is input quality.
Put agents in channels deliberately. An agent added to a channel reads that channel; that is both the value and the exposure.
Use the surfaces for their purpose rather than defaulting everything into one stream. Fast working chat and considered async writing benefit from different rooms, and agents behave differently in each.
Do not import the Slack habit of many low-signal channels. Each one accumulates context that dilutes rather than compounds.
If you ignore this
Used well, a Buzz workspace gets better over months: context accumulates, agents get more useful, and the audit trail becomes a genuine record of how a project was reasoned through.
Used as a chat tool, it gets worse: sprawling channels produce agents that confuse projects, and the latency cost of the relay buys nothing. The failure is gradual and hard to attribute, which is why the mental model matters at the start rather than at the point of frustration.
Examples
Slack is an office where you can call an AI helpline. Buzz is an office where some of your colleagues happen to be AI — they sit in the meetings, read the room, and are accountable for what they do in it.
An analysis channel holding three weeks of project conversation, with an analyst agent as a member, produces materially better analysis than the same agent given the same question in a fresh chat — because it has been in the room.
A general #ai channel where everyone asks unrelated questions of a shared agent. Nothing accumulates, the context is noise, and the agent performs worse than a plain chat assistant would.
Verified against Buzz v0.5.20 · Updated 2 Sep 2026