E7 · Swarm and peer network
Family: Multi-agent
Every agent can talk to every other; with no central coordinator, the path emerges.
Principle
Every agent can communicate with every other. There is no coordinator: the execution path is emergent and takes shape through the exchanges.
The problem it solves
You want to avoid a single central point, bottleneck or point of failure, and let agents organise themselves on a fuzzy problem.
When to use it
- Ambiguous problems that benefit from debate, for those who accept a high cost and a hard-to-bound result (Google Cloud).
When to avoid it
- Almost everywhere today: swarms of parallel writers still see no meaningful adoption (Cognition).
- The risk of non-convergence is real and errors amplify: prefer a central orchestrator (E1).
Common pitfalls
The anti-patterns that misuse this pattern, best established first. Each sheet describes the symptoms, the public incidents and the remedies.
Loop, stopping and orchestration
- AP35 · Unbounded agent loop — Nothing in the code stops the agent: no turn cap, no budget, no deadline.
- AP39 · Unchecked propagation — One agent’s output becomes the next one’s input unchecked, and a local error ends up in the deliverable.
- AP40 · Parallel writers — Several agents edit the same deliverable at once, each with its own implicit decisions.
Synonyms by vendor
- Swarm (Google Cloud)
- Network (LangGraph)
- Decentralized multi-agent collaboration (AWS)
- Horizontal multi-agent (IBM)
- Decentralized and independent topologies (Kim et al.)
Neighbouring patterns
Sources
- Towards a Science of Scaling Agent Systems — Y. Kim et al. (Google, MIT), arXiv, 2025-12-09
- Choose a design pattern for your agentic AI system — Google Cloud Architecture Center, 2026-05-28
- Multi-Agents: What’s Actually Working — Cognition (W. Yan), 2026-04-22
Content reviewed on 2026-10-02