PodBrowser
Startup Ideas

Jack Dorsey's Buzz: The New Hermes Agent?

Tuesday, 28 July 2026 · 2 min read · Listen to the episode ↗

Jack Dorsey's Buzz is a team collaboration tool built on the open Nostr protocol that positions itself as a competitor to both Slack and GitHub by storing all chats, agent interactions, and data on user-controlled relays rather than locking data inside a proprietary platform. Agents are first-class team members in Buzz, with a swappable harness supporting Claude Code, Codex, and Goose without losing conversation history, and tight Git integration lets agents spin up parallel work trees simultaneously.

Jack Dorsey launched Buzz, a team collaboration tool built on the open Nostr protocol, and it is being positioned as a direct competitor to both Slack and GitHub. The most important distinction from Slack is not the surface similarity but the openness: all chats, agent interactions, and data are stored on relays that users can self-host, whereas Slack controls all user data and makes migration extremely difficult.

Agents are first-class team members in Buzz rather than bolt-on integrations, and the agent harness is fully swappable between tools including Claude Code, Codex, and Goose without losing existing chat context or conversation history. Skills and skill files stored in a global directory remain accessible to agents after a model switch. One analyst runs two named agents, Fizz and Honey, pinned to different models based on task complexity to manage token consumption, with Fizz using a Fable model and Honey using Sonnet.

Buzz has tight Git integration that allows agents to create feature branches and work in parallel work trees without modifying files locally on the user's machine. Agents can spin up multiple parallel versions of a task simultaneously, such as three different landing page designs generated from a single chat discussion. Buzz offers its own hosted git service alongside GitHub integration, which is the basis for the claim that it is competing with GitHub as well as Slack.

A concrete demonstration involved building a full-stack CRM app using the WASP framework, deploying it live to Railway end to end without detailed technical instruction, and then sharing the deployed app with team members who joined the chat and collaboratively directed agents to add or change features. A separate workflow involved building a tweet leaderboard app and configuring an agent named Fizz to check the app's public API daily and deliver stats and top tweets into a Buzz channel automatically, replacing a manual process of copying data into ChatGPT or Claude.

Buzz includes audio huddle functionality for live voice calls with agents that receive the full context of the existing chat, though this feature was not personally tested and listeners were directed to YouTube demonstrations instead. Bitcoin Lightning payments are not yet integrated but are expected, with a prediction that agents will be able to pay for compute or users will be able to tip one another for completed tasks via Lightning. Buzz also has a shared compute feature allowing users to download local AI models and share them across multiple devices from a single machine.

Buzz is early preview or possibly alpha software, and recurring workflows did not function reliably during testing. It operates more slowly than working directly in Claude Code or Codex because of relay server communication overhead. It is considered best suited currently for solopreneurs and small teams iterating on small software projects rather than complex engineering work, and the analyst is attempting to convince his own team to adopt it as a replacement for Discord-based communication.

This summary was generated from the episode transcript and can contain mistakes.