Plan with PRDs and delegate tasks to cheaper agents

A developer uses a high-level agent to write a PRD, splits it into a task list, and hands the work to cheaper agents in a production workflow.

What They Did

The commenter described an internal setup with two systems: a task-based workflow that enforces TDD and can run multi-agent flows with git trees, plus persistent memory with hooks and house-style rules. Their day-to-day process is to work with a high-level agent like Opus to create a PRD, break it into a task list, and then hand implementation work to cheaper agents; they also said they are building their own Python orchestrator to call Haiku or other agents by API.

How It Went

The source gives no hard performance numbers or business results. It frames the approach as an actively evolving internal workflow rather than a finished system, with the user still building the harness around it.

Worth Knowing

The practical catch is that this depends on custom infrastructure the commenter created, including task enforcement, persistent memory, and an in-progress orchestrator.