Multi-agent systems / orchestration / agent swarms

Agent Chimera

A sourced field guide to the engineering decision behind multi-agent AI: when a team of specialized agents is stronger than one agent, and when it is just a more expensive way to lose context.

A lead agent connected to research, builder, verifier, policy, memory, and handoff agents.
The chimera metaphor only works when the parts share enough state to act like one organism.

The thesis

Multi-agent is a task shape, not a maturity badge.

A multi-agent system is useful when separate agents can pursue different parts of a problem with their own context, tools, or evaluation criteria, then return artifacts that a coordinator can verify and synthesize.

That is why breadth-first research, independent review, translation across languages, and policy-separated handoffs are plausible fits. It is also why tightly coupled implementation work, small support flows, and high-risk actions often belong in one agent, a deterministic workflow, or a human approval loop.

"finding the simplest solution possible"
Source: Building effective agents

Map

The field guide map

How to use this site

Start with single vs multi if you are making an architecture call. Use patterns when you already know a split is justified and need vocabulary. Use failure modes before you ship anything that lets agents call tools or hand work to one another.

Every page links back to primary sources and includes a "Cite this page" block. The stance is deliberately balanced: the same site that cites Anthropic's strong research result also keeps Cognition's context-fragmentation warning in view.

Read next