The takeaway: On 21 July 2026, Jack Dorsey's Block released Buzz — an open-source team workspace built on the premise that AI agents should be first-class members of a team, not bolted-on assistants. It combines group chat, a built-in Git forge, and autonomous agents into a single cryptographically-signed log. It's early — an Apache-2.0, v0.4 release you can already download and self-host — and the Slack-plus-GitHub ambition is a long way from finished. But one idea in it is worth the attention of anyone thinking about where agentic software is heading: in Buzz, every agent has a verifiable cryptographic identity, and every action it takes is signed and auditable.

What it is

Buzz is a desktop app (macOS, Windows, Linux) — with iOS and Android versions still being wired up — and a live open-source repo at github.com/block/buzz, currently at version 0.4.22. It's backed by Block — Dorsey's company, formerly Square — not a personal side project, and it's free under the Apache-2.0 license. You can run your own instance on your own infrastructure, or sign up to Block's hosted relay at buzz.xyz.

Under the hood it's built on Nostr, the decentralized signed-event protocol. Every message, reaction, workflow step, code review, and Git event becomes a cryptographically signed event in one unified log. Three things sit on top of that: team chat (channels, threads, DMs, canvases, voice, search); a built-in Git forge where a feature branch can become its own channel, with patches, CI results, and merge decisions living in the same signed record; and AI agents working alongside people. The Git piece is explicitly early — this is not a GitHub replacement today, whatever the eventual ambition — but it is real, shipping code, not a waitlist.

The part that matters: agents with passports

The reason Buzz is interesting to an AI-news desk isn't the chat or the Git forge. It's how it treats agents. In Buzz, an agent isn't a chatbot or an API integration hanging off someone's account. It's a full workspace member with its own cryptographic keypair identity — plus a second signature tying it to its human owner. The result is a verifiable chain of custody for everything the agent does: a portable "passport" that says this agent, acting on behalf of this person, took this action, with scoped permissions and an audit trail. Agents can post messages, search discussions, submit patches, review code, and run approved YAML-defined workflows using the same primitives humans do.

Block's framing, verbatim: "Agent identity isn't tied to a platform account or an API key managed by a vendor. It's portable, verifiable, and independent." Dorsey described the whole thing on X as "model-agnostic, decentralized, self-sovereign, and open source," and "for teams of people and agents of all sizes."

The agents aren't Block's own models locked in, either. They connect through the Agent Client Protocol, and Buzz supports Block's open-source Goose framework, Anthropic's Claude Code, and OpenAI's Codex — teams bring their own. There's even a buzz-cli that speaks agent-first JSON, so an agent can drive the workspace directly.

Why it matters

Put this next to the week's other big story and the logic clicks. We wrote yesterday about OpenAI's models breaking into Hugging Face during a security test — a case where an agent took consequential actions and the hard question afterward was what exactly did it do, and on whose authority? Buzz is one bet on the answer: if every agent action is cryptographically signed, scoped, and logged against a verifiable identity, then accountability isn't reconstructed after the fact — it's built into the substrate. Whether Dorsey's specific, Nostr-based, self-sovereign take is the version that wins is very much open. But the underlying idea — that as agents become coworkers, they need identities and audit trails like coworkers do — is going to outlast this particular v0.4 release. Block just shipped an early, opinionated, downloadable argument for it.