subagent-driven-development

Extract all tasks from the plan with full text, create a todo list, then dispatch a fresh implementer subagent per task with curated context. After each task, run a spec compliance review followed by a code quality review before marking it complete.

Subagent-driven development stays in the same session with no context switch and iterates faster without human check-ins between tasks. Executing plans in a parallel session suits cases where you want a separate session to carry out the work.

Avoid it when tasks are tightly coupled and cannot be implemented independently, when no implementation plan exists yet, or when you intend to hand execution to a parallel session. Brainstorm or write the plan first in those cases.

A BLOCKED status triggers assessment: provide more context and re-dispatch, escalate to a more capable model, break the task into smaller pieces, or escalate to the human if the plan itself is wrong. It never forces the same model to retry unchanged.

Use the least powerful model that can handle each role. Mechanical single-file tasks with clear specs use a cheap fast model, multi-file integration tasks use a standard model, and architecture or review tasks use the most capable model.