Parallelism speeds software delivery by doing independent work at the same time. How much work can proceed in parallel depends on feature dependencies: dependent tasks may still allow their plans or reviews to happen alongside implementation. section 6.12 section 6.9
Choose the scale of parallel work to fit the task:
- Delegate separable work to subagents within one Claude Code session; each runs in its own context and returns its result to the main session. section 6.4
- Run separate Claude Code instances on independent features when dependencies allow. section 6.9
- When feature slices must be built in sequence, generate or review the next slice's plan while the current slice is being built. section 6.9
Delegate test runs
Ask Claude Code: “Run the tests using a subagent for each component and report back with the summary.” section 6.4
Figure 6.2 Subagents run as separate sessions with their own context. They’re still subject to the permission mode and when they finish they send a summary back to the main agent.
Shows subagents as separate sessions with their own context, subject to the permission mode, that return a summary to the main agent. section 6.4






















































