subagent-driven-development

Read the plan once, extract all tasks with full text into a todo list, then dispatch a fresh implementer subagent per task with the complete task text and 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, uses a fresh subagent per task, and iterates faster without human-in-the-loop between tasks. Executing plans in a parallel session involves a handoff to a separate session.

Avoid it when there is no implementation plan yet, when tasks are tightly coupled and cannot be delegated independently, or when you need a parallel session instead of same-session execution. In those cases, brainstorm or write a plan first.

Assess the blocker: provide more context and re-dispatch if it is a context problem, use a more capable model if it needs more reasoning, break the task into smaller pieces if too large, or escalate to the human if the plan itself is wrong. Never force the same model to retry unchanged.

Spec compliance confirms the implementer built exactly what was requested, nothing missing or extra. Reviewing code quality first wastes effort on code that may need rework to meet the spec, so quality review only starts after spec compliance passes.

Use a fast, cheap model for mechanical tasks touching one or two files with a complete spec, a standard model for multi-file integration work, and the most capable model for architecture, design, and review tasks.