Parallel Work for Claude Codes
Parallel work in Claude Code uses subagents and background agents to split a job. A subagent runs in its own context with a focused prompt and tool set, returns a summary and keeps the main conversation small, while background agents continue long tasks so you can work on something else.
Stage: Extending · Part of the Claude Code roadmap · Last reviewed 19 November 2026
What to learn in Parallel Work
- Define a subagent with a clear purpose and limited tools
- Delegate a codebase search to a subagent to save context
- Run independent tasks in parallel and merge the results
- Decide which tasks are too coupled to parallelise
- Review a subagent's summary critically before relying on it
Parallel Work interview questions
Why use a subagent instead of doing everything in one session?
A subagent has a separate context window, so exploring many files or logs does not fill the main conversation. It returns only a summary. This keeps the main session focused and lets different tasks use specialised instructions or tools.
When is parallel agent work a bad idea?
When tasks edit the same files or depend on each other's output, because results conflict or need rework. Parallelism suits independent work like reviewing separate modules or researching different questions, ideally in isolated branches or worktrees.
Where Parallel Work fits in the Claude Code roadmap
Parallel Work belongs to the Extending stage (beginner level), which takes about 6 hours. Tools and agents.
Other units in this stage
Back to the full Claude Code roadmap