Agent Lore

When to parallelize

Use parallel agents when tasks are independent and don't share state:

• Code review on one file + linting another
• Searching two different directories
• Building frontend + documenting API

In Claude Code

Use the Task tool to launch concurrent sub-agents:

Launch two agents in parallel:

Agent 1: Explore src/components/ and list all React components
Agent 2: Explore src/api/ and list all API endpoints

Both agents run simultaneously and return their results.

In Cursor

Open multiple Composer tabs or use the ask feature in parallel panels. Each operates with its own context.

Anti-patterns

• Launching agents that depend on each other's output (sequential, not parallel)
• Spawning one agent per file in a 50-file refactor (use a shell loop instead)
• Parallelizing trivial tasks (the overhead isn't worth it)

Rule of thumb

If two tasks require reading different files and producing independent outputs, parallelize. If task B needs task A's result, run them sequentially.