CH 11 · Subagents and Multi-Agent Orchestration · 01/11
Read as one pageSubagents and Multi-Agent Orchestration
A subagent gets its own context window so noisy research does not fill the main conversation. Covers orchestrator and verifier patterns.
11 slides · 6 min read
In this chapter
By now you have a single agent that loops, builds prompts, manages context, runs tools safely, and remembers things. The next leap is to let it spawn other agents. A subagent is a fresh agent instance with its own context window that the parent hands a task; it works autonomously and returns a result. This is the "multi-agent" layer (layer 5) and "agent contracts and orchestration" (pillar 4).
The single most important reason subagents exist is context isolation, and it follows directly from Chapter 4. A research or exploration task generates a huge amount of noisy intermediate output (file dumps, search results, DOM snapshots). If that all lands in your main conversation, it eats your context window and your budget. A subagent does that messy work in its own window and returns only a tidy summary. Cursor's docs say it plainly: "Long research or exploration tasks don't consume space in your main conversation." The parent stays focused on decisions; the subagent absorbs the mess.
A real-life analogy: you are leading a project and need a competitor analysis. You do not read fifty articles yourself and clutter your desk; you ask a researcher to do it and bring back a two-page brief. You get the conclusion without the fifty tabs. The subagent is that researcher, and its fresh context window is the separate desk.
Finished this idea?
Your learning progress stays in this browser.