subagent-driven-development

Subagent-driven development is an agentic coding workflow that improves plan execution by dispatching a fresh implementer subagent per task to prevent context pollution, followed by mandatory spec compliance and code quality review gates.

To enforce spec compliance and code quality gates, execute tasks sequentially and run two review stages after implementation, forcing fix-and-retry loops on reviewer findings until the code passes both spec verification and quality checks.

No, you cannot run implementation subagents in parallel because this workflow strictly prevents parallel execution to avoid conflicts, ensuring each task is isolated and completed within the same context before the next begins.

You should use isolated subagents to prevent context pollution across independent tasks, ensuring each implementation starts fresh, which maintains correctness and maintainability while preserving the broader plan context.

Subagent-driven development is designed for plan-driven sessions where tasks are mostly independent and executed within the same context, making it less suitable for tightly coupled tasks that require shared mutable state across implementations.

The limitation of two-stage review gates in plan-driven development is the strict sequential processing overhead, requiring spec compliance approval before code quality review and blocking subsequent tasks until current fix-and-retry loops are fully resolved.