What is an Agentic Harness: Buzz and Human-Agent Collaboration

Four days ago we started with Hermes. Hermes is the toolbox and management infrastructure you need to run an agent. Yesterday we saw OpenRig—managing coding agents like Claude Code and Codex as a team with mid-level orchestration. But both these harnesses are fundamentally inward-facing: a single machine, a dev environment, or a closed system.

But what if you asked this: What if humans and agents worked in the same workspace, reading each other’s outputs, building on each other’s work, exchanging ideas in the same thread?

Buzz is the answer to exactly that question.

The Social Dimension of the Harness

Hermes manages a single agent. OpenRig manages agent team coordination within coding workflows. Buzz goes further—it’s a protocol where human and agent organisms live in the same social network.

Picture this scenario:

It’s morning. I’m building a product at an early startup. The team has four humans: me, a product manager, a designer, and a reviewer. On the same team are two agents: Claude Code writing software, and a Codex instance doing code review.

Open the workspace. Five humans and two agents, all in channel 1, thread 1.

Product manager: “We have a timeout issue in the user login screen.” Designer: “I checked the design too, CSS isn’t causing the problem.” Claude Code agent: “Let me get details.” (It frames the issue, starts writing code.) Human developer: “I checked the repo, want more detailed logs?” Codex agent: “Code saved and reviewed. Found a performance issue.” (Reviews and approves.) Reviewer: “Operations tracked. Approval came back. Ready for production.”

Everyone in the same space. Everyone can see everyone else’s work. Everyone in the same context—not digging through saved logs.

This is Buzz’s core idea. Agents see each other and humans. Who is working, who approved, what decisions were made—all in one session, in a verifiable chain.

What is Buzz, Technically

Buzz is an open source collaboration platform released by Block under the Apache 2.0 license. Its architecture is built on the Nostr protocol—a decentralized, cryptographic identity and audit log network.

Buzz’s architecture splits into two parts:

Relay (Central Workspace): Channels, direct messages, workflows, a git forge, chat logs, and search. You can host this yourself. The relay records all events in a signed event log. No event can be forged, no identity can be fake.

Agent Harness (buzz-acp): A client that connects to the relay over WebSocket, runs agents, and hosts any LLM-based agent like Claude Code, Codex, or Goose. The harness starts agent work, sends output to the relay, and receives commands from humans. Agents run on the machine where your code lives, with their own tools and access.

This split gives a critical advantage: the relay doesn’t manage heavy computation. The architecture is lightweight and scalable. Agents run in their own environment. The relay acts as the coordinator.

Why This is a New Idea

Slack and GitHub have treated agents as an afterthought. Agent channels open, bot messages send occasional updates, but the agent is not really a team member.

In Buzz, an agent is a team member from day one. It has identity, permissions, signed messages, and history. A human can review an agent’s work, and an agent can review another agent’s. Deadlocks, disagreements, approvals—all recorded in session.

Also, agents can swap harnesses. If you use Claude Code today, you can move to Codex tomorrow. All conversation, git history, decision history travels with you. You’re not locked into a vendor.

Buzz in Production

Real scenario: a product company building a new feature. Buzz relay runs on their own server. Two coding agents, one review agent, four humans. Working in the same channel for a week. Agents complete features, open pull requests, human team members review, agents add review comments, all within the same space.

At week’s end, looking at the product history: every event, every decision, every signature, all there. No Slack log purge. No lost context.

Buzz and Agentic Organization

Buzz is a new arrival in the agentic harness category. Unlike Hermes or OpenRig, Buzz adds a social layer. But all three—Hermes, OpenRig, Buzz—solve the same core problem: turning agents from tools into a coordinated system.

Yet each harness targets a different problem. Hermes manages a single agent. OpenRig coordinates agent teams in coding workflows. Buzz structures human-agent team collaboration in a trustworthy, auditable way.

These open source harnesses open a door for companies: integrate agents into product work, but with your own control, your own hosting, your own choice. You can look at Buzz and other harnesses, but if corporate alignment, data ownership, and truly advanced agent governance matter, agentic-organization, as a file-based, self-hosted platform, is built to manage these harnesses within organizational schema. agentic-organization (github.com/fab-agent/agentic-organization) recognizes agents as first-class organizational members and manages them in organization configuration. Open source harnesses run agents; organizational platforms govern them.

You can try Buzz today. It’s open, free, runs on your own server. Excellent as a team starting point.