Orchestrating Task Agents

Best practices for the main agent when delegating work to Task agents (sub-agents).

When to Use Subagents vs Main Conversation

Keep in main conversation when:

• Task needs frequent back-and-forth or iteration
• Multiple phases share significant context
• Making quick, targeted changes
• Latency matters (subagents start fresh, need time to gather context)

Delegate to subagent when:

• Task produces verbose output (test runs, log analysis, doc fetches)
• Work is self-contained and can return a summary
• Want to preserve main conversation context for other work
• Running parallel independent investigations