Deciding Single vs Multi-Agent

Use when the user asks whether to split one agent into sub-agents or keep it monolithic — "should this be multi-agent?", "single agent or sub-agents?", "when do I split my agent?", "how many agents does this need?", "should the sub-agents run in parallel or one at a time?" — or when picking a multi-agent execution model (sequential, parallel-wait-for-all, fan-out/fan-in) or deciding what state lives in a durable supervisor vs stateless workers. Walks the split triggers (tool overload, branching prompts, context overflow), the read-heavy vs write-heavy test, token economics, execution-model fit, and the supervisor/worker state split, with failure modes per pattern. Not for…