TL;DR — Key Takeaways
- Block’s Buzz treats AI agents as full workspace participants, not chatbots attached to existing tools.
- Each agent receives a cryptographic identity, defined permissions and a signed record of its actions.
- The platform is designed to make agent identities more portable and less dependent on vendor-issued tokens.
Block, the fintech company formerly known as Square and co-founded by Jack Dorsey, has released Buzz, an open-source workspace where employees and AI agents operate side by side. The project moves past the idea of AI as a chat assistant bolted onto existing tools. Instead, agents show up as full participants, each with a cryptographic identity, a defined scope of permissions, and the ability to post messages, review code, and run workflows alongside people.
Buzz is built on Nostr, the decentralized protocol better known for censorship-resistant social media. That choice turns out to be the interesting part of the story. Every message, reaction, workflow step, review approval, and Git event in Buzz is a signed event in one log, giving the same shape, identity model, and audit trail whether the author is a person or a process. In practical terms, an AI agent working inside Buzz doesn’t get a special side-channel or a shadow account. It gets a keypair, just like a human teammate would, and its actions land in the same searchable history.
Block describes Buzz as a self-hostable workspace where humans and AI agents share the same rooms, and the project leans into that framing hard. Agents get their own surface area through tools like buzz-cli, an agent-first command-line interface, along with an ACP harness that connects to coding agents such as Goose, Codex, and Claude Code, plus YAML-based workflow automation and persona packs. On the human side, there’s a desktop app built with Tauri and React. Under the hood, the relay runs on Rust, with Postgres storing events, Redis handling pub/sub, and Typesense powering search.
What Buzz is really trying to solve is something most enterprise AI tools have quietly ignored: portable, vendor-independent identity for agents. Right now, most agent deployments tie an AI system’s identity to whatever platform issued it — an OAuth token from one vendor, an API key from another. That works until you want to move the agent, audit it independently, or scope its permissions the way you’d scope a new hire. Buzz’s pitch is that identity should belong to the agent itself, signed and portable, not locked to whichever platform happens to be hosting it that week.
Mitch Ashley, VP and Practice Lead for Software Lifecycle Engineering & AI-Native Software Engineering at The Futurum Group, sees that keypair model as the real story here. “Buzz’s real move is giving agents a keypair instead of a platform’s OAuth token,” Ashley said. “That’s the difference between an identity you audit and scope like a new hire versus one you’re stuck trusting whichever vendor issued it.”
That intersects with a problem security teams have been raising all year. Non-human identities — service accounts, API keys, and now AI agents — already outnumber human users in most enterprises, and by a wide margin. One recent industry report puts the ratio at more than 80 non-human identities for every human in the average enterprise, with the total machine identity count roughly quintupling since 2021. Governance hasn’t caught up. One survey found that 68% of organizations can’t reliably tell AI agent activity apart from human activity on their own systems. Ashley argues that’s exactly why the decision needs to happen at the protocol layer rather than inside any one vendor’s console. “Identity governance decided at the protocol level travels with the agent instead of locking to one vendor’s stack,” he said. “Enterprises piloting agentic tools should demand that portability before they’re faced with navigating around it.”
Buzz is still early. Block’s own project notes are candid about what works today versus what’s still being wired up — the relay, channels, and agent CLI are functional, but things like mobile clients and workflow approval gates are in progress, and reputation systems across relays remain aspirational. The project explicitly asks people not to plan compliance programs around the features that don’t exist yet. That’s a reasonable caveat for anyone evaluating Buzz for production use today.
Block released Buzz under an Apache 2.0 license, and it’s already being built in public — the README notes it was drafted inside a Buzz channel by agents and a human working together. Whether Nostr becomes the identity layer other agentic platforms adopt, or just an interesting experiment from a company known for taking unconventional bets, Buzz is a useful signal of where the industry’s thinking is headed: agents as accountable participants in the workspace, not just tools that respond when called.
