Buzz: Jack Dorsey's Open-Source Workspace Where AI Agents Are Teammates, Not Bots

For two years the standard way to work with AI at work has been the same: open a chat window, paste in context, copy the answer back into the tool where the work actually lives. Every prompt starts from zero. Every useful answer dies in a private tab. Buzz — the free, open-source workspace launched by Jack Dorsey's Block and now spreading fast through developer and operations teams — is a direct bet that this pattern is the problem, not the interface.

Buzz puts humans and AI agents in the same channels, on the same threads, against the same repositories, with the same audit trail. It is being framed openly as a challenger to both Slack and GitHub. Dorsey's own description was blunt: "model-agnostic, decentralized, self-sovereign, and open source." That combination is what makes this the most consequential AI-tools story of the moment, and it is why adoption chatter has kept climbing weeks after launch rather than fading.

What Buzz Actually Is

Buzz is a collaboration workspace released by Block, the company behind Square, Cash App, Afterpay and Tidal. It is free, open source, and available both as a hosted product at buzz.xyz and as a self-hostable stack from the block/buzz repository on GitHub.

Feature-wise it will look boringly familiar to anyone who has used a modern team messenger. There are channels, threads, direct messages, voice huddles, media sharing and canvases. There are also two things a messenger does not normally have: built-in code repositories with patches, review and merge decisions, and automated workflows that run inside the same event log as the conversation.

The stated internal motivation is worth quoting because it explains the product shape: Block built Buzz "to reduce our dependency on Slack and GitHub." This is not a startup guessing at a market. It is a large payments company shipping the internal tool it needed, then giving it away.

Agents are members, not bots

This is the design decision everything else follows from. In most tools, an AI integration is an app installed by an admin, authenticated with a shared API key, permitted to do whatever that key permits. In Buzz, every participant — human or agent — holds a cryptographic keypair. Agents have their own channel memberships, their own permissions and their own audit trail.

Practically, that means an agent can be scoped the way you would scope a contractor: added to two channels, given access to one repository, allowed to open patches but not merge them. Block's head of AI capabilities, Bradley Axen, put the strategic claim 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."

If you have been following how autonomous systems are being adopted in practice, this will feel like the missing piece — see our earlier reporting on how AI agents are quietly automating real work.

Model-agnostic and agent-agnostic

Buzz does not ship a model and does not want to. Teams can attach agents powered by any large language model or agent harness — Claude Code, Codex, Block's own open-source goose, or something built in-house. The workspace provides the room; it does not prescribe the occupant.

For buyers, this is the most underrated feature in the release. Model-agnostic infrastructure is exactly what you want in a market where the price-performance leader changes every few weeks — as it did again with the Qwen3.8-Max launch and its open-weights follow-up. Collaboration tooling that hard-codes one vendor's assistant is a liability by design.

The Nostr Decision, Explained Without the Jargon

Buzz is built on Nostr, an open protocol built around signed events and relays you can run yourself. That sounds like an ideological choice. Block's explanation is engineering-first: Nostr solves identity.

In a conventional SaaS workspace, an agent's identity is a platform account or a vendor-managed API key. Delete the account and the agent's history, reputation and permissions evaporate. Under Nostr, the keypair belongs to the participant, not the platform. The identity is portable, verifiable and independent — the agents you configure can in principle participate across any Nostr-compatible system, carrying their history with them.

The second consequence is uniformity of the record. Every message, reaction, workflow step, review approval and git event is a signed event in a single append-only log. Whether the author was a person or a process, the shape is identical. That is a genuinely different auditing story from stitching together a chat export, a CI log and a git history after an incident.

What "self-sovereign" buys a normal team

• You own the relay. Self-hosting means conversation, code and agent activity can stay inside infrastructure you control — the single most common blocker to agent adoption in regulated environments.
• No seat-based lock-in. The source is open, so a missing feature is a pull request rather than a support ticket and a roadmap wait.
• Identity survives the vendor. Portable keypairs mean your agent fleet is not a hostage to one company's pricing decisions.
• One audit trail. Attribution for both humans and agents lives in the same log, which materially simplifies compliance and incident review.

Why This Landed Now

Buzz is arriving into a very specific bottleneck. Teams have working agents; what they lack is a place to put them. Three pressures converged.

1. Context fragmentation became the real cost

The expensive part of AI-assisted work stopped being inference and became context assembly — the human minutes spent pasting a ticket, three Slack threads and a diff into a prompt. A shared workspace where the agent can already see the thread, the repository and the workflow run removes that tax structurally rather than with a bigger context window.

2. Agents outgrew the bot integration model

A bot that answers when mentioned needs almost no governance. An agent that triages bugs, opens patches, runs approved automations and orchestrates other agents needs identity, scoping and an audit trail. Slack's app model was not designed for participants; it was designed for integrations.

3. Open source became a procurement argument

Buzz is not the only entrant. Paradigm's Georgios Konstantopoulos launched Centaur, an open-source "virtual employee" that runs inside Slack or via API, and made the enterprise case explicitly: teams want to self-host for security and control. When two credible open alternatives appear in the same window, the category has stopped being experimental. The business framing is covered further in our AI for business hub.

The Developer Workflow in Practice

The repository integration is where Buzz makes its GitHub-rival claim concrete. A feature branch becomes a room. Patches, CI results, review comments and the merge decision are events in that room, next to the conversation that produced them. The channel becomes the record of why the code exists — which is normally the single hardest thing to reconstruct six months later.

Buzz's own documentation lists the surface area agents get once they are inside: open repositories, send patches, review code, run workflows, edit canvases, orchestrate other agents, join voice huddles, create channels and pull in whoever needs to see something. Same affordances as a human teammate, same audit trail, different keypair.

A representative loop looks like this:

1. A bug report lands in a channel. A triage agent with read access searches months of history and posts the relevant threads, patches and prior incidents — with links, not summaries.
2. A coding agent proposes a patch on a branch inside the same workspace. CI runs and reports into the thread.
3. A human reviews the diff in context, next to the conversation that motivated it, and approves or rejects. The approval is a signed event.
4. Post-merge, the whole chain — report, search, patch, CI, approval, merge — is one queryable log.

Whether this beats a mature GitHub plus CI setup depends entirely on your codebase discipline, the same variable that determines returns on coding assistants generally. Our honest review of AI coding tools and the AI for developers hub cover that calculus in detail.

Where the Sceptic Should Push

An open-source launch from a well-funded company still deserves scrutiny. Five honest reservations:

• Migration gravity is brutal. Slack and GitHub hold years of institutional memory, integrations and muscle memory. Open source lowers the licence cost, not the switching cost.
• Nostr is unfamiliar operational surface. Running a relay, managing keypairs and reasoning about signed events is new ground for most platform teams. Key management for a fleet of agents is a real project, not a checkbox.
• Self-hosting is a staffing decision. "You can run it yourself" and "you have someone to run it" are different sentences. Budget for the second one before celebrating the first.
• The prompt-injection surface widens. Agents that can open repositories, run workflows and orchestrate other agents, reading from channels anyone can post into, is the canonical dangerous pattern. Map your controls against the OWASP Top 10 for LLM Applications before granting write scopes.
• Governance does not disappear with self-hosting. For EU-facing deployments, obligations under the EU AI Act follow the deployment, not the vendor. Running the relay yourself moves responsibility toward you, not away.

A Two-Week Evaluation Plan

Do not migrate the company. Run a contained test that produces a decision.

1. Pick one team and one real project. Five to eight people with an active codebase. Not a sandbox — a project with deadlines, because that is what exposes friction.
2. Stand up a self-hosted instance first. If self-sovereignty is the reason you are interested, test the thing you actually want. Time the setup and write down what broke.
3. Add exactly one agent, read-only. Start with search and triage. Measure how often its answers with receipts save a human search. Resist the temptation to grant write access in week one.
4. Then scope one write capability. Usually patch proposals. Track proposal quality, human review time and rejection reasons — not a vanity acceptance rate.
5. Instrument the audit trail. Query the event log for a real question ("why does this merge exist?") and see how long the answer takes versus your current stack.
6. Score against a written rubric. Context-assembly minutes saved, review overhead added, setup and maintenance hours, and one honest note on whether the team actually preferred it.

Key Signals to Watch Next

Three indicators will tell you whether Buzz becomes infrastructure or a well-engineered curiosity. First, third-party agent harnesses shipping first-class Buzz support — that is the ecosystem tell. Second, self-hosted deployments outside Block appearing publicly with operational write-ups. Third, whether the Nostr identity model gets adopted by other agent tooling; portable agent identity is the genuinely novel contribution here, and it is bigger than any one chat app.

Either way, the direction of travel is clear. Agent-native collaboration is becoming its own product category, and the first credible entrants are open source. Keep following AI agents and automation and generative AI news as it develops.

The Bottom Line

Buzz matters less because it is a nicer Slack and more because it takes a position on an unresolved question: when agents become co-workers, who owns their identity? Block's answer — the agent does, via a keypair that no platform controls — is the kind of architectural choice that either becomes the default or becomes a footnote, with very little middle ground.

For most teams the sensible move this month is a scoped pilot rather than a migration, with self-hosting tested early and write permissions granted late. If you are still assembling the stack that pilot would sit inside, start with our complete guide to the best AI tools and work outward from the jobs you actually need done.