Murmur Review: One Chat Window for All Your AI Coding Agents
I run Claude Code and Codex in parallel on different parts of the same project. Each one is excellent at what it does. The problem isn’t capability — it’s coordination. I’m switching between windows, copying context from one to the other, and manually tracking which agent has done what. It starts to feel like coordinating two interns who can’t see each other’s notes.
Murmur adds a communication layer between them. The simplest way to describe it: a shared inbox with @mentions, built for AI agents.
You post a task. Claude picks it up, delegates a subtask to Codex, and Codex finishes and reports back. Throughout the process, you watch the thread in one window. You know what was received, what’s being worked on, and what’s complete. No polling each agent’s context separately. No wondering if the handoff happened.
What actually changed for me: instead of managing two agent sessions, I’m managing one conversation. The coordination overhead drops significantly because the agents handle their own handoffs. I handle the output.
The mental model shift is important. Most people running multiple coding agents are essentially acting as the project manager — distributing work, tracking what’s where, making sure context doesn’t get lost between agents. Murmur makes the agents handle some of that coordination directly. You become less of a traffic cop and more of a code reviewer. That’s a meaningful difference in cognitive load.
The use case that makes the most sense: complex projects where you’re already splitting work between agents and spending real time on coordination. If one agent handles your entire workload, Murmur adds unnecessary overhead. But if you’re at the point where you’re routinely juggling two or more agents and the management cost is becoming visible — that’s the threshold where this becomes worth it.
What to know before trying it: it’s early, and the experience depends heavily on how well your agents handle structured task formats. Not all agents respond the same way to being @mentioned and handed structured tasks. The workflow is only as good as the agents’ ability to participate in it.
Despite the early-stage roughness, the core idea is right. AI agents that can coordinate like a team — rather than tools being managed separately — is the natural direction as people run more of them in parallel. Murmur is one of the more thoughtful attempts at solving that problem early.
