subagent-driven-development

Read the plan once, extract all tasks with full text into a todo list, then dispatch an implementer subagent per task followed by spec compliance and code quality reviewer subagents. Continue through all tasks without pausing, then run a final whole-implementation review.

Subagent-driven development stays in the same session with fresh subagent context per task and no human-in-loop between tasks. Executing plans in a parallel session involves a context switch and handoff, which suits work you want isolated from your current session.

Avoid it when tasks are tightly coupled and cannot be split independently, when no implementation plan exists yet, or when you prefer a parallel session handoff. Brainstorm or write the plan first if requirements are still unclear.

Assess the blocker: provide more context and re-dispatch for context problems, re-dispatch with a more capable model for reasoning-heavy tasks, break oversized tasks into smaller pieces, or escalate to the human if the plan itself is wrong. Never force the same model to retry unchanged.

Use a fast, cheap model for mechanical tasks touching 1-2 files with complete specs, a standard model for multi-file integration work, and the most capable model for architecture, design, and review tasks. Match model power to task complexity signals.