Subagents / agent swarm

A coordination pattern in which one agent delegates bounded work to other agents that have their own run context, tools, or instructions. Swarm is a looser label for larger or more distributed multi-agent systems.

A coordinating agent can delegate security analysis, performance analysis, and style checks to separate runs, or assign one agent to gather evidence that another later uses. Separate contexts reduce direct context competition, but they do not guarantee focus, file isolation, independence, or useful diversity. Agents may share the same model and blind spots, and their tasks may be parallel, sequential, or mutually checking.

Builder example

Subagents give you speed through parallel execution and focus through scoped context. The cost is coordination complexity: subagents can produce conflicting results, duplicate work, or fail silently while the main agent continues without noticing. Every subagent also adds token costs and makes the overall system harder to debug, since you need to trace through multiple agents to reconstruct what happened.

Common confusion: More agents does not mean a smarter or more independent system. Shared training and prompts can create correlated errors, while unclear ownership creates duplicate edits and contradictory outputs.