Peer-to-Peer Pattern

What is the peer-to-peer pattern?

In the supervisor pattern, a "manager" sits on top calling the shots. The peer-to-peer pattern flips that: a group of agents hold equal rank, nobody is in charge, and they talk to each other directly — sending messages, negotiating division of labor, and cooperating to hit a shared goal, like colleagues huddled around a whiteboard.

How is it different from the supervisor pattern?

Supervisor: a pyramid
One central node coordinates everything, with a clear hierarchy — but if that node fails, the whole thing suffers.
Peer-to-peer: a mesh
No single point of control. If one agent goes down, the others keep going, so it's more fault-tolerant.

How do the agents cooperate?

Direct messaging
Agents send task requests and results to each other — whoever needs something just asks.
Negotiation and voting
On a disagreement, they can negotiate or even vote on which path to take.
Publish and subscribe
One agent publishes a message, and the agents that care subscribe to receive it.

Strengths and challenges

The upside is decentralization — more robust, more flexible. The challenge is coordination overhead: with no "commander", you can get duplicated effort or agents waiting on each other, and splitting work and merging results need real design.

Bottom line: the peer-to-peer pattern is a group of agents as equals, getting things done by talking it out with each other.