We're already drowning in AI-generated garbage PRs, and now Jack Dorsey wants to give these things cryptographic identities and full access to our Git repos? Here's the problem: Buzz, Block's new open-source workspace, launched yesterday, July 21, 2026, and it's trying to solve a problem we haven't even fully defined yet. The idea is to combine team chat, AI agents, and Git hosting, all built on Nostr. On paper, "self-sovereign" and "decentralized" sound great. In practice, we're looking at a potential data privacy minefield with a single point of failure.
The pitch is simple: ditch Slack and GitHub, run your own Buzz relay, keep your data. It's written in Rust, which is a solid choice for systems-level stuff, and it uses Nostr events, cryptographically signed, for everything from messages to code changes. This means every action, every approval, every line of code, gets a durable, portable identity and an audit trail. That's the cool part. Your human employees and your AI agents share this identity structure, complete with key pairs and channel memberships.
But here's where the rubber meets the road. Buzz treats AI agents as "first-class" users, not just some bolted-on integration. They get their own identities, can be granted access to specific channels, and have an internal authentication and permission protocol. These agents can search discussions, open repositories, submit patches, review code, run workflows, edit shared canvases, and even create channels. Block provides harnesses for Goose, Codex, and Claude Code, keeping the model choice separate.
This "first-class" agent concept is a massive blast radius waiting to happen. Think about it: an AI agent, with its own cryptographic identity, can submit patches and review code. What happens when that agent hallucinates a critical security vulnerability fix that actually introduces a backdoor? Or, more likely, just submits a patch that doesn't compile? (I've seen PRs this week that don't even compile because the bot hallucinated a library). The audit log is great for after the fact, but it doesn't prevent the initial screw-up. The complexity of managing access rulesets for these "multiplayer agents" to prevent accidental disclosure or "career ruining messages" is going to be a nightmare.
The Git integration is baked in, using standard Git Smart HTTP. Feature branches can become dedicated channels, and patches, CI results, review comments, and merge decisions are all preserved within the same record, sharing a single search index. This sounds like a unified developer experience, but it also means a single point of compromise could expose your entire development history, discussions, and code.
Now, about that "decentralized" claim. Buzz is an open-source, self-hostable Nostr relay. Organizations can run their own, keeping domain and data control with portable Nostr key pairs. That's the promise. The current reality, however, is a dealbreaker for true decentralization:
- No peer-to-peer event exchange.
- No gossip layer.
- No replication between relays.
All reads and writes within a workspace pass through a single relay. This relay acts as the authoritative server for authentication, signature verification, event storage, and update distribution. So, while you own the relay, it's still a centralized choke point for your workspace. If that relay goes down, or gets compromised, your "decentralized" workspace grinds to a halt. This isn't decentralized; it's just self-hosted. It's a single point of failure with extra steps.
The project is labeled "unfinished" by Block's own documentation. Mobile clients, push notifications, and workflow approval gates are still pending. Version 0.4.21, released yesterday, includes fixes, but the core architecture still has these glaring limitations.
My take is this: Buzz is an interesting experiment, but it's not ready for prime time. The "first-class" AI agent concept, without solid, battle-tested guardrails and a truly decentralized architecture, is a liability. The current "decentralized control, centralized relays" model is a misnomer; it's just self-hosting with Nostr identity. Until they sort out the peer-to-peer event exchange and replication, you're just moving your single point of failure from Slack's servers to your own. That's not progress; it's just a different kind of risk. Focus on stability, not just features.
