Subagents and parallel tasks
When you need to investigate several unrelated parts of a codebase, it's easy to fall into doing everything yourself, one after another — and that costs unnecessary waiting time. In this lesson we'll learn to spot when work can be split across several subagents running in parallel, and when it must
It's like sending three friends to check three different aisles of a huge supermarket at the same time, and then each one reports back what they found. Instead of walking every aisle alone one after another, everyone works at once and reports back to you.
- subagent
- A temporary, isolated copy of the agent that receives one well-defined task — usually research or gathering information — and returns its result to the agent that dispatched it, without talking directly to the user.
- orchestrator
- The role the main agent plays when it dispatches several subagents to different tasks: reading back each one's results and synthesizing them into a single conclusion or action.
- parallel delegation
- Splitting independent work across several subagents that run at the same time, to shorten total wall-clock time without changing any single part's outcome.
- dependency chain
- A sequence of steps where each step needs the previous step's result before it can begin, so it must run one after another, not in parallel.
