Buzz, Jack Dorsey's Group Chat for AI Agents, Explained

Buzz is a new app from Jack Dorsey's Block that puts coding agents like Claude Code and Codex into channels where you message them the way you would message a team, and that shift from single-model tools to multi-agent chat is the core idea worth paying attention to. The video is a hype-heavy first look. The creator says he now reaches for Buzz more than Codex and frames it as the "endgame" for agents. Underneath the overclaims, the actual product is real, open source, and does something specific: it removes provider lock-in and lets several agents work in one conversation. This post covers what Buzz is, what the video gets right, and where it oversells.

What Buzz Actually Is

Buzz is an open-source workspace from Block, the company Jack Dorsey runs. It launched on July 21, 2026. The frames in the video show the product page with the tagline "Where people and agents work together" and a "Developer Preview" label, plus a "Get the app" button and a GitHub link. The design is a bright yellow Slack-style layout with a bee mascot.

The core function is a group chat where AI coding agents live in channels as members. You message an agent, it responds, and multiple agents can collaborate in the same thread. Buzz also folds in Git repositories, so chat, code, and agents sit in one window. Block released it under the Apache 2.0 license with the source on GitHub, and it runs on macOS, Windows, and Linux.

The Real Advantage: No Provider Lock-In

The strongest point in the video is about lock-in. The creator contrasts Buzz with OpenClaw, where "you had to use one thing as a brain and then you had to use all these other models." In Buzz you can drop in Claude Code, Codex, or Block's own Goose agent, and you are not tied to a single provider acting as the controller.

This is accurate. Buzz is model-agnostic and built on the open Agent Client Protocol. You bring your own models and point the agents at whatever LLM you want. The creator garbles one detail: he says "kimmy k3," which does not match Buzz's supported agents. The third first-party option is Goose, Block's own open-source agent framework, not a Kimi model. The general claim holds even though that specific name is wrong.

The "Group Chat for Agents" Framing

One frame shows a post from Riley Brown describing the experience: messaging Codex, Claude Code, and Cursor from a phone all morning, and it "feels like a group chat, and they can even collaborate." That captures the pitch better than the video's own narration. The novel part is not any single agent. It is putting several agents in one conversation and letting them hand work to each other.

Whether this is genuinely more productive than running agents in separate terminal sessions is an open question the video does not answer. "Feels like a group chat" is a UX observation, not evidence of better output. Treat it as an interface bet, not a proven workflow upgrade.

Where the Video Oversells

The framing is the weakest part. Calling Buzz "the endgame for agents" and declaring that "it's over for anything that is trying to compete in that space" is standard tech-influencer hyperbole. Buzz launched as a developer preview in mid-2026. It has not displaced anything. Slack, GitHub, Cursor, and the terminal-based agent tools all still exist and all still have users.

The "OpenClaw is dead" claim is also asserted, not demonstrated. The video gives no usage numbers, no migration data, and no comparison beyond the creator's own habits. His personal preference for Buzz over Codex is a data point of one. The underlying product is interesting on its own merits. It does not need the funeral announcements, and you should discount them.

Key Takeaways

Buzz is real and open source. It is a Block product, launched July 21, 2026, under Apache 2.0, with the code on GitHub and desktop apps for macOS, Windows, and Linux.
The core feature is multi-agent chat. Claude Code, Codex, and Goose can all work in the same channels and collaborate, which is the genuine novelty.
No provider lock-in is the real selling point. Buzz is model-agnostic via the Agent Client Protocol, so you bring your own models rather than being tied to one controller.
The video gets a detail wrong. "Kimmy k3" appears to be a garbled reference. Buzz's third first-party agent is Block's Goose, not a Kimi model.
Discount the "endgame" language. Declaring OpenClaw dead and the space "over" is unsupported hype. Buzz is a promising developer preview, not a finished category winner.