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.
Done withClaude CodeClaude
- 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.
Try it yourself with Claude Code
I want to build [feature or project] in [repo path]. First write a short PRD covering goals, scope, and acceptance criteria, then split it into small, independent tasks in a TASKS.md file with clear done criteria for each. Stop and show me the PRD and task list so I can review it before anything is implemented, so I can hand the tasks to a cheaper model.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: Reddit · Apr 1, 2026
Related use cases
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Build prospect lists and cold emails from the web
Claude Code driving the author's logged-in Chrome read seed pages and LinkedIn profiles to find 25-30 sourced prospects; about 20 emails sent drew 7 replies.
Summarize non-obvious methods file by file in a paper project
A commenter had Claude Code write file-by-file summaries of non-obvious methods in a research project, then reviewed them to keep track of the work.
Review Linux kernel patches
Google engineers launched Sashiko, an agentic AI code reviewer for the Linux kernel.
Custom agent
Refactor a million-line repo with 1,393 subagents
A Nous Research developer had Hermes dispatch 1,393 subagents over about 19 hours to clean up its own repo, cutting non-test Python source by 34.4% in a merged PR.
Crawl documentation into clean Markdown for AI agents
The author built a documentation crawler for their AI agents that pulled 60 pages of clean Markdown in 49 seconds.
Custom agent
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.