First look: Buzz, where the agents sit in the channel

Buzz is the new one from Jack Dorsey's Block, launched late July 2026. The pitch is Slack territory with a twist that matters to me: AI agents aren't a feature bolted onto the chat, they're members. They sit in the channel next to the humans, and the platform's own framing folds Git projects into the same window. It's open source, model-agnostic, and built on Nostr, an open protocol, which is the basis of its self-sovereignty claim: the workspace is meant to outlive any vendor. Block admits it's early days, and it shows. That's fine. Early is when the architecture is easiest to read.

The reason I'm paying attention is that I already run a fleet, and my fleet coordinates the unfashionable way: shared files under version control, append-only ledgers, and a messaging front I keep deliberately dumb, a place for pointers and nothing else. Buzz is a bet in the opposite direction. It says the chat room itself is the coordination surface. The conversation is the workspace.

What I'm actually evaluating

Not the features. The claim underneath them. Three questions I want the hands-on time to answer.

• Where does durable state live when the log is a conversation? A chat scrolls. My whole memory architecture is built on the idea that state you can't audit is state you don't have. If an agent's decisions live in channel history, I want to know what happens when you need to reconstruct them six weeks later.
• Does an agent-in-the-channel sharpen accountability or blur it? In my system, one writer owns an artifact and everything else advises. A room full of members editing the same work is exactly where that discipline goes to die, unless the platform enforces it better than a prompt would.
• The self-sovereign angle is the part most aligned with how I build. I rent intelligence and own memory. A workspace on an open protocol, where the data outlives the vendor, is the same instinct applied to the team layer. If Buzz gets that right, it's the most interesting thing about it.

Status: reading the architecture and standing it up next to the fleet to see what breaks. No verdict yet. That's what makes it a note and not an essay. I'll report what the hands-on time actually shows, including the boring answer, if the boring answer is the true one.