Delegate with subagents

🎯Goal. Offload noisy side-work (searching, log-reading, test runs) to a subagent so your main conversation stays focused.

1. 1A subagent runs in its own context window with its own system prompt and tool access. When a side task would flood your main chat with search results or logs you won't reread, delegate it — only the summary returns.
2. 2Define a custom subagent when you keep spawning the same kind of worker; give it a clear description so Claude knows when to delegate, and route simple tasks to a cheaper model (e.g. model: haiku).
3. 3Use subagents to isolate high-volume operations — running a big test suite or processing a log file — keeping the verbose output out of your main context.

✓You'll see. A research- or test-heavy step done in a subagent, returning a short summary instead of pages of output in your main thread.

💳Cost. Subagents control cost by keeping verbose output in their own window and by routing simple work to faster, cheaper models like Haiku.

💡Takeaway. When a task would bury your conversation in output you won't reuse, hand it to a subagent and get back just the answer.