Claude Code can run multiple agents at the same time — each one working on a separate task in parallel.
You have 2+ tasks that are:
- Independent — fixing one doesn't affect the other
- Non-overlapping — they don't write to the same files
Classic examples:
- 3 test files failing for different reasons → one agent per file
- 4 subsystems to refactor → one agent per subsystem
- Multiple API integrations to build → one agent per integration
You (coordinator)
├── Agent A: Fix auth tests
├── Agent B: Fix payment tests
└── Agent C: Fix notification tests
All three run at the same time. You review the results when they're done.
Open Claude Code and run tasks in parallel by creating multiple TaskCreate entries:
Task("Fix the authentication test failures in tests/auth.test.ts")
Task("Fix the payment processing failures in tests/payments.test.ts")
Task("Fix the notification failures in tests/notifications.test.ts")
- Break your work into independent units
- Open multiple Claude Code sessions (one per task) or use the Agent tool
- Each agent gets a focused, self-contained prompt — no shared context
- You review each result and integrate
Each agent gets exactly the context it needs — nothing more.
Bad: "Here's our whole codebase, please fix everything"
Good: "Fix the failing test in tests/auth.test.ts. The error is: [paste error]. The relevant files are: [list them]."
A focused agent with complete context outperforms a general agent with too much context every time.
- Don't parallelize tasks that share state (same file, same database table, same config)
- Don't parallelize when step 2 depends on step 1's output
- Don't send the same broad prompt to multiple agents — each should have a different scope
For the full parallel-agents skill used to orchestrate this in Claude Code: → Install the Superpowers skill pack: https://get.caffier-agency.com?utm_source=instagram&utm_medium=gist&utm_campaign=PARALLEL