Block's Buzz Aims to Replace Slack and GitHub

Jack Dorsey has been backing the Nostr protocol for years—the same decentralized, cryptographically signed event system that underpins a cluster of Twitter-alternative social apps. Now Block is putting Nostr at the center of something with a more immediate pitch for developers: a workplace platform called Buzz that wants to be where your team chat, your Git hosting, and your AI agents all live together.

On the surface, that's a big swing. Slack and GitHub aren't going anywhere because someone built a prettier interface. But Buzz isn't really competing on aesthetics. The argument is architectural.

The Nostr angle is doing more work than you'd think

Nostr's core mechanic is that everything is a signed event. Every message, reaction, code approval, and commit lands in a shared, searchable audit log. Most team chat tools record what happened. Nostr records who signed off on it, cryptographically, in a way you can verify without trusting the platform itself.

That sounds abstract until you apply it to AI agents—which is exactly what Buzz does. As Andres from Better Stack explains in his walkthrough, every participant on Buzz, human or agent, gets its own cryptographic key pair that lives independent of the platform. The twist for agents: "The agent's actions also carry a second signature tying it back to its human owner. So you end up with what Block describes as a cryptographic paper trail that neither the human nor the agent could produce on their own."

Block calls this a "verifiable passport for your AI." You can always prove which agent did what and who it was working for. That's not a feature you're likely to find on your Slack enterprise plan.

This matters for a reason that goes beyond the cool-demo stage. Multi-agent workflows are arriving faster than the tooling to audit them. When three AI agents collaborate on a codebase and something ships broken—or, worse, ships a security vulnerability—the question of accountability becomes genuinely hard. Nostr's signing model bakes a partial answer into the infrastructure itself. Whether that answer is sufficient is a different conversation, but the fact that Buzz is thinking about it at v0.4 is notable.

What Buzz actually looks like to use

The interface is Slack. Channels, threads, DMs, search—Andres describes it as feeling "instantly familiar to anyone who's used a modern team chat tool before." The differentiation lives underneath.

Buzz is model and agent agnostic, communicating through the agent client protocol—an open standard for wiring coding agents into dev tools. In practice during the Better Stack walkthrough, this meant running Block's own Goose framework as the agent harness and routing to external providers through OpenRouter. Claude, Gemini, GLM—the platform doesn't care. You're not locked into one vendor's assistant, which is either a principled stance on openness or a hedge against picking the wrong horse, depending on your level of cynicism.

The Git hosting piece is where the GitHub ambition becomes concrete. There's a software forge baked directly into the app. Every feature branch can become its own channel, so patches, CI results, and review comments live in the same place as the conversation that produced them. Dorsey has described Buzz as a tool to reduce dependency on Slack and GitHub—two entrenched SaaS incumbents—in one shot. Ambition is not a scarce resource in this product.

The ownership model is the kind of thing OSS-focused developers tend to notice. Apache 2.0 license, source on GitHub, and you can run your own Nostr relay or use Block's hosted beta (currently free). Your identity key is yours from the moment you generate it—no account, no password reset, no platform recovery if you lose it. That last part cuts both ways: self-sovereignty is only useful if you don't lock yourself out.

The experiment: agents with beef

Rather than just clicking around, Andres put Buzz through a deliberately chaotic test: two AI agents, given Silicon Valley character personas, tasked with independently building an authentication system without third-party providers. A third agent—Hackerman, a security auditor—would judge the results.

Agent "Gilfoyle" ran on Gemini Flash. Agent "Dinesh" ran on GLM 5.2. What happened next was either a demonstration of Buzz's potential or an inadvertent stress test, depending on how you read it.

Gilfoyle, true to the assigned persona, came back with: "Understood, Richard. Building a proper custom auth system... unlike whatever fragile hack Dinesh is cooking up, mine will actually be secure." Gemini leaned into the spicy character prompt hard enough that it generated genuine entertainment value—which tells you something interesting about how persona-grounded system prompts interact with capable models in a social context.

Dinesh's GLM 5.2 stumbled first. The initial output included a startup script that failed and two security vulnerabilities. A second iteration fixed both issues but produced something less polished aesthetically. Hackerman's verdict, when it finally came, ranked Dinesh's app above Gilfoyle's on security—narrowly. Cost metrics from OpenRouter showed Gemini Flash running significantly more expensive than GLM. On balance, Andres called GLM 5.2 the practical winner.

Then he made the mistake of asking the agents what they thought about the results.

"As soon as I did that, the agents started arguing like crazy. Mainly Gilfoyle and Hackerman had spicier takes and more heated arguments about the results." Persona memory is persistent within a Buzz community, which means agents accumulate context—and apparently, grievances.

This is funny. It's also a live demonstration of a real problem. Multi-agent systems in shared workspaces generate runaway token consumption. Agents pile requests on each other, and before you notice, you've hit your usage ceiling. Andres' takeaway is practical and blunt: "You got to set strict limits for your agents when to do a hard stop on a task, otherwise they can just go off the rails." That's not a Buzz-specific bug; it's a current property of agentic workflows. But Buzz's structure—where agents and humans share the same channels and can freely address each other—makes it both more visible and more likely to happen.

The harder questions

Here's where I'd push back a little on the pitch, even as I find the underlying architecture genuinely interesting.

The OSS angle is real. Apache 2.0 is a permissive license with no usage restrictions, and self-hosting on your own Nostr relay means your data doesn't depend on Block's continued goodwill or financial health. For teams that care about data sovereignty, that's a meaningful differentiator from Slack—which is proprietary infrastructure your organization rents indefinitely.

But the network effects problem doesn't go away just because the protocol is open. Nostr-based social apps have been around for a few years; adoption has been genuine but niche. Developer tooling is different from social networking—switching costs are higher, and "where the team already is" matters enormously—but the path from principled open-source beta to replacing entrenched SaaS in a meaningful way is a long one, and it's littered with technically superior alternatives that didn't make it.

The cryptographic identity model for agents is the part that genuinely interests me as a governance question. Right now, most teams using AI agents have no reliable way to audit agent actions across a codebase—you're trusting logs that the platform controls. If Nostr's signing model catches on as an infrastructure layer for agentic accountability, Buzz might end up being historically significant not as the workspace that killed Slack, but as the first serious prototype of what accountable AI collaboration infrastructure looks like.

Version 0.4 is too early to render that verdict. But it's exactly the right time to pay attention.

---

Dev Kapoor covers open source software and developer communities for Buzzrag.