Block Buzz: AI Agents Get a Cryptographic Identity (2026)

On July 21, 2026, Jack Dorsey's Block released Buzz, a free, open-source workspace where people and AI agents share the same channels, code, and workflows.¹² The pitch that separates it from every "add a bot to Slack" integration is small but structural: in Buzz, each agent gets its own cryptographic identity.

Instead of running under a shared service account or a borrowed API key, an agent holds its own keypair — plus a second signature that ties it back to the human who authorized it.³ Every action it takes is signed and traceable. Block is betting that this one primitive is what the agent era actually needs.

In one line: Block released Buzz, a free, open-source workspace where humans and AI agents collaborate as equals. Its defining idea is that every agent gets its own cryptographic identity plus a signature linking it to the human who authorized it — creating a tamper-evident chain of custody that API-key integrations can't provide.¹³

TL;DR

• What launched: Buzz, an open-source (Apache-2.0) collaboration workspace from Block, built on the Nostr protocol. It shipped July 21, 2026 as an early, pre-1.0 desktop release for macOS, Windows, and Linux; mobile apps are still pending.¹³
• The core idea: Every participant — human or agent — holds a secp256k1 keypair. Agents get their own identity plus a second signature binding them to a human owner, so every message, commit, and workflow step is a signed, tamper-evident event.³
• Why it matters now: About 57% of organizations already run agents in production, per LangChain's 2026 survey, yet 68% say they can't reliably tell an AI agent's actions apart from a human's, per a Cloud Security Alliance study.⁴⁵
• What runs in it: Model- and agent-agnostic. Ships with harnesses for Claude Code, OpenAI Codex, and goose, plus a built-in Git forge so teams can host repositories without GitHub.¹⁶
• The catch: It's early software with no announced business model, entering a market where Slack and Microsoft Teams are deeply entrenched — and Block itself is mid-restructuring after a Q1 net loss.³⁷

What You'll Learn

• What Buzz is and exactly what shipped on July 21
• How Buzz gives each AI agent a cryptographic identity and a chain of custody
• Why service accounts and API keys can't answer "who authorized this agent?"
• What running on Nostr does — and doesn't — mean (it isn't peer-to-peer)
• Which agents run inside Buzz, and how Block already uses them at scale
• Whether replacing Slack and GitHub is realistic, and what's still missing

What Block Buzz actually is

Buzz is a collaboration workspace that looks familiar on the surface. It has channels, threads, direct messages, voice, media sharing, code repositories, and automated workflows — the shape of any modern team chat tool.¹

The difference is who counts as a participant. In Buzz, AI agents aren't assistants that answer when tagged. They hold their own identities and permissions, and they post, review code, run approved automations, and join conversations alongside humans.¹

Dorsey framed the launch bluntly on X: "a new groupchat platform for teams of people and agents of all sizes, built to reduce our dependency on slack and github. Model-agnostic, decentralized, self-sovereign, and open source."³

Block published Buzz under an Apache-2.0 license at github.com/block/buzz. Teams can self-host their own instance, or sign up for Block-hosted infrastructure at buzz.xyz.¹⁸

"Every company is going to need a place where humans and agents work together," said Bradley Axen, Block's Head of AI Capabilities. "The question is whether that place is proprietary or open. We built Buzz because we believe it should be open."¹

How Buzz gives every AI agent a cryptographic identity

Here's the mechanism. Every participant in Buzz — person or agent — holds a secp256k1 keypair, the same elliptic-curve cryptography that underpins Bitcoin.³

Humans get one keypair. An agent gets its own keypair, plus a second cryptographic signature that ties it back to its human owner.³

That pairing creates a verifiable chain of custody. Block's engineering team describes the design plainly: "The agent's owner signs a narrowly scoped authorization. The agent then signs its own work with its own identity."⁹ So when an agent merges a pull request or triggers a workflow, the signed record shows two things at once: that the agent acted, and which human authorized it to act.⁹

This matters because the identity belongs to the agent, not to the platform. It isn't a vendor-managed API key or a login tied to one company's account. It's portable and verifiable, and it can travel with the agent across any Nostr-compatible system.¹

The audit trail falls out of the design rather than being bolted on. Every action — a message, a reaction, a commit, a CI result, a review comment, a merge approval — is stored as a signed event, so the log is unified, searchable, and tamper-evident.³

Why an API key was never enough

The question Buzz is really answering is a governance one: which human authorized this agent to do this, and when?

Today, most enterprise agents run under a service account, a user's API key, or session credentials that expire and regenerate. When something goes wrong — a bad commit, an unauthorized call, a workflow that spends money it shouldn't — tracing responsibility back through those layers is a forensic exercise.³

The scale of that gap is measurable. A 2026 Cloud Security Alliance study, commissioned by Aembit, found that 68% of organizations cannot clearly distinguish AI agent activity from human activity in their systems.⁵

And this isn't a fringe problem. LangChain's 2026 State of Agent Engineering report found roughly 57% of surveyed organizations already run agents in production.⁴ The agents are doing real work with real access; the accountability layer just hasn't caught up.

A signed keypair changes the default. Unlike an API key — a bearer token that works for anyone holding it — a cryptographic identity proves the specific agent acted, and the owner-signature says who stood behind it.

Table: How agent identity and accountability differ between a conventional API-key integration and an agent in Buzz. Source: NerdLevelTech analysis of Block's Buzz announcement, engineering blog, and repository docs, July 2026.

Built on Nostr: what "decentralized" really means here

Buzz runs on Nostr — short for "Notes and Other Stuff Transmitted by Relays" — an open protocol created in 2020 and best known for censorship-resistant social apps. Dorsey has backed Nostr financially for years.³

In Nostr, every event is a small, signed JSON object — it carries an id (a hash of the content), the sender's public key, an event kind, arbitrary tags, and a Schnorr signature. In Buzz, those events cover every action in the workspace.³

One architectural nuance matters if you're evaluating it. Buzz is not peer-to-peer. Each deployment runs on a single authoritative relay.³

So "decentralized" here means any organization can run its own relay and keep full control of its data, configuration, and agents — not that there's no central server within a given deployment.³ Block also runs a hosted relay, free during the beta.¹

Self-hosting is real but not trivial. The quick-start path needs Docker and Hermit — or a manual toolchain of Rust 1.88+, Node 24+, pnpm 10+, and the just task runner.¹⁰ Notably, Block says it specified Buzz's Git-on-object-storage protocol in TLA+ and model-checked it for durability, reconstruction, and concurrent pushes — an unusual level of formal rigor for such early software.⁹

The agents (and the Git forge) inside Buzz

Buzz is model-agnostic and agent-agnostic. It ships with harnesses for three coding agents: Anthropic's Claude Code, OpenAI's Codex, and goose — the terminal-first coding agent Block open-sourced in January 2025 and later contributed to the Linux Foundation's Agentic AI Foundation.¹¹¹

Those harnesses connect through the Agent Client Protocol (ACP) — Block says any agent that "speaks the Agent Client Protocol" can work in Buzz.⁹ ACP is the open standard Zed Industries introduced in August 2025 to wire agents into developer tools the way LSP standardized language servers for editors.¹²

Buzz also includes a built-in Git forge, so teams can host repositories inside the workspace rather than on a separate host. Feature branches become channels, and patches, CI results, and review comments land as signed Nostr events beside the discussion that produced them.⁹ Block notes the Git hosting backend is still being wired up.¹

Block is its own first customer: the team says it built Buzz inside Buzz, even drafting the launch write-up "in a Buzz channel with our team and our agents."⁹

Block already runs agents at scale, too. Its internal BuilderBot system logs 200,000 operations a day, merges about 1,500 pull requests a week, and accounts for roughly 15% of the company's production code, per figures Block posted on X.¹³

There's an irony worth noting: BuilderBot is orchestrated from Slack today, with engineers tagging it in a thread.¹³ Buzz is Block eating its own dogfood to move that work off Slack and into a workspace it controls.

Replace Slack, or live inside it?

Buzz reflects one side of a genuine strategic fork in how AI agents enter the workplace. Block's bet is to replace the incumbent tools — a Nostr-native workspace built for agents from the protocol layer up.

The other side is to upgrade the incumbent. In May 2026, Paradigm and Tempo open-sourced Centaur, a "multiplayer, self-hosted, secure" agent that lives inside Slack or runs via an API.¹⁴ Paradigm's Georgios Konstantopoulos has argued that the near-term opportunity is embedding capable agents in the tools teams already use, not asking them to switch wholesale.³

The incumbents are formidable. Slack and Microsoft Teams together dominate enterprise messaging — Teams counts hundreds of millions of monthly active users, and Slack reaches a large majority of the Fortune 100 by third-party estimates.³ Distribution like that isn't overcome by a better identity model overnight.

Block also isn't launching from a position of financial strength. The company reported a Q1 2026 net loss of about $308.7 million, reversing a roughly $190 million profit a year earlier, amid AI-driven restructuring and a workforce reduction of more than 40%.⁷ Buzz enters as a free, Apache-licensed product with no announced commercial trajectory.

What Buzz can't do yet

This early in its life, Buzz is infrastructure Block has published openly, not a polished product. Mobile apps are unfinished, push notifications are pending, and some workflow features lack complete execution paths.¹⁰

The self-hosted path takes engineering time — standing up a Nostr relay, managing Rust and Node versions, and wiring in existing databases and codebases before the first channel exists.³

That narrows the immediate audience to software teams already coordinating AI agents across a codebase and willing to run their own infrastructure. Solo operators and non-technical teams aren't the target, and Buzz doesn't pretend otherwise.³

The open-source model also defers the question every open-core enterprise tool eventually faces: what pays for it? Block says the hosted relay is free during beta. It hasn't said what comes after.³

The Bottom Line

Strip away the Slack-versus-GitHub headline and Buzz is an argument about a single missing primitive: identity. As agents move from experiments to production — roughly 57% of organizations already run them⁴ — the tooling built for human-only teams has no clean way to answer who authorized an agent to do something.

Buzz's answer is to give every agent its own keypair and bind it to a human with a second signature, so accountability is legible by default instead of reconstructed after an incident. That's a genuinely novel foundation, and Block is already running its own agents against it.⁹

Whether it dislodges anyone is a different question. Buzz is early, unmonetized, and up against a Slack–Teams duopoly, from a company mid-restructuring.⁷ The identity idea may well outlast the app — the more durable claim here isn't that Buzz replaces Slack, but that "which human authorized this agent, and when" is about to become a requirement, not a nice-to-have.

Working with AI agents that take real actions? Related reading: treating AI agents as insider threats, Microsoft's Agent 365 control plane, and agent-to-agent identity in the A2A protocol.