On July 21, Block — the fintech company led by Jack Dorsey, formerly Square — released Buzz, an open-source collaboration platform where humans and AI agents work in the same workspace as peers, not through a chat window bolted onto existing tools. It’s free, Apache-2.0 licensed, available hosted at buzz.xyz or self-hosted from github.com/block/buzz.
The framing from Block is deliberately blunt: “the most productive work doesn’t happen when someone asks AI for help. It happens when humans and agents are in the same room, working on the same thing, with shared context.” That’s the pitch — not another assistant panel, but a shared space where an agent’s presence looks structurally the same as a teammate’s.
What’s actually in it
Buzz bundles the collaboration surface you’d expect — channels, threads, direct messages, voice, media sharing — with code repositories and automated workflows in the same place. Two design choices separate it from a Slack-plus-bot integration:
Agents carry cryptographic identities with defined permissions, rather than acting as an API call triggered from a human’s account. The platform is model-agnostic — Block says it supports Claude Code, Codex, Goose, and custom agents — plus integrations with databases, CRMs, codebases, and file systems, with Git integration still described as early-stage.
The identity layer is built on the Nostr protocol, the same decentralized identity/relay system used for censorship-resistant social apps. That choice is the actual thesis of the product: an agent’s identity is a keypair, not a session token issued by whichever platform it’s running on. Bradley Axen, Block’s Head of AI Capabilities, put the stakes plainly: “Every company is going to need a place where humans and agents work together. The question is whether that place is proprietary or open.”
Why the identity model is the interesting part
Most “agent platforms” today authenticate agents the same way they authenticate humans clicking through OAuth — a token scoped to a session, owned by the platform, gone the moment you switch providers or the platform changes its terms. If an agent’s entire identity and permission history lives inside one vendor’s database, migrating to a different model or a different tool means starting the trust and audit trail over from zero.
A portable, protocol-level identity is a different bet: the same cryptographic key could in principle persist across tools, so an agent’s collaboration history and permission grants aren’t tied to a single company’s continued goodwill or uptime. That’s a genuinely different answer to a problem every team running multiple agents against multiple providers will eventually hit — vendor lock-in, but for infrastructure rather than data.
It’s worth being precise about what that portability does and doesn’t cover. A durable identity key doesn’t make an agent’s behavior portable — swap the underlying model and the agent will act differently even if its cryptographic identity and permission grants haven’t changed. What Nostr-based identity buys you is a durable record of who did what and the ability to keep that record outside any one company’s servers, not a guarantee that your agents behave the same way after a model change.
The open question Block is actually asking
Buzz launched a few weeks ago, self-hosted and Apache-2.0 licensed, which means the honest read right now is “reference architecture,” not “proven production tool.” Block hasn’t published detailed guidance yet on how the platform handles multi-agent conflict resolution, content-safety filtering for agent-directed instructions inside shared channels, or operational tooling like PR governance and observability — the kind of scaffolding that turns a protocol demo into something a team can run in production without building the rest themselves.
That’s not a knock on the idea. It’s the normal state of a brand-new open-source release, and the fact that it’s forkable under Apache-2.0 means the gaps are addressable by anyone, not just Block. The strategic question Axen is posing — proprietary or open — is the one worth sitting with regardless of how mature the current implementation is: if your team is already wiring multiple agent frameworks into shared tooling, do you want the identity and permission layer underneath that work to be something you can inspect, fork, and run independently of any single vendor’s roadmap?
Related
- Your AI Agent Is a Financial Liability
- Goose AI: A Free, Local Alternative to Your $200/Month Claude Code Bill
Buzz is early. But it’s the first serious attempt to answer the identity question for multi-agent teams with a protocol instead of a platform — worth watching whether frameworks like AutoGPT or LangChain start treating Nostr-backed identity as infrastructure, or whether it stays a Block experiment.