Block launched Buzz on July 21, 2026. It is a free, open-source workspace where AI agents join your team as members, not as chatbot sidebars. Each agent holds its own cryptographic key pair. Every action it takes is signed and traceable back to the human who authorized it. The workspace runs on the Nostr protocol, model-agnostic, decentralized, and published under Apache 2.0.

This matters because it names a problem most companies have not articulated yet: their AI works alone in a private window and nobody else on the team sees what it does.

The Problem Nobody Named

Right now, AI usage in most organizations looks like this: one person opens a chat interface, types a prompt, gets a response, and copies something into a document or Slack thread. The rest of the team has no visibility into what the AI produced, what context it had, or whether its output was good. There is no shared accountability. No audit trail. No way for a colleague to build on what the agent already did.

Gallup reported on July 21, 2026, that 52% of US employees now use AI at work. The majority use it for rewriting text and running searches. Solo. In a window nobody else sees.

This is the chatbot model. It treats AI as a private tool. You use it the way you use a calculator. And like a calculator, its work disappears the moment you close the window.

Block built the opposite of that.

What Buzz Actually Does

Buzz looks like Slack or Teams on the surface. Channels, direct messages, voice, code, workflows. The difference is who is in the room.

AI agents join channels as members. They have their own names, their own permissions, their own activity streams. When an agent reviews code, kicks off an automation, or posts a summary, the entire team sees it happen in real time. Not buried in a sidebar. Not hidden behind a one-on-one conversation. In the shared space, alongside everyone else’s work.

Jack Dorsey described it as “a new groupchat platform for teams of people and agents of all sizes.” That framing is deliberate. Not a tool platform. Not an automation layer. A groupchat. The thing your team already uses to coordinate work. Except now agents are in it.

The cryptographic identity layer is what makes this more than a chat app with a bot integration. Every agent holds a key pair. Every message it sends carries two signatures: the agent’s own, and a second one that ties it to the human who authorized it. If an agent takes an action, you can trace exactly what it did, when, and who gave it permission.

That is not a feature for engineers. That is a feature for the director who needs to explain to their VP what the AI on their team actually does all day.

Why This Is Not a Bot Integration

Slack has bots. Teams has Copilot. Every collaboration tool on the market has bolted AI onto the side of the product in some form. The difference is the mental model.

A bot integration treats AI as a servant. You summon it with a slash command. It responds. It goes away. It has no persistent presence, no ongoing context, no accountability that the rest of the team can see.

Buzz treats AI agents as participants. They have a continuous presence in the workspace. They build context over time. Other team members can see their work stream the same way they see a colleague’s activity. And because the identity is cryptographic, the accountability is baked into the protocol, not dependent on a vendor’s logging system.

This distinction sounds philosophical until you try to scale AI usage past one or two people on a team. The moment you have three people using AI independently, you hit a coordination problem. Duplicate work. Conflicting outputs. No one knows what the agent already handled. The private-chat model breaks down the same way email breaks down when a project outgrows two people.

What This Means for Your Team

If you run a team that uses AI today, ask one question: can anyone on your team see what the AI produced for someone else without asking them to forward it?

If the answer is no, you are running AI the way companies ran email in 1995. One person at a time. No shared context. No compounding.

What separates teams over the next twelve months will not be model quality or subscription tier. It will be whether the AI’s work is visible to the rest of the team — shared, accountable, and building on itself.

Block did not invent this idea. But they built the first open-source, protocol-level implementation of it. And they gave it away for free.

The private chatbot window was always going to be temporary. Agents as teammates, in the room, doing work everyone can see — that is the architecture this is heading toward. Block just shipped the first real version of it. The interesting question is not whether this model wins. It is how long the private-window holdouts take to notice.