Cole Medin · Cole Medin · 18:38.
Keep the core loop
- Plan a release with a person, split it into dependency-aware tickets, then run plan → implement → validate → review.
- Merge passing work into staging, review the release as a whole, and feed problems back as issues.
Add shared management as teams grow
- Version skills and agents centrally, with owners, permissions and approval status.
- Bring service ownership and business context into planning through Port’s MCP server.
- Use repository scorecards to track readiness and propose repairs when checks fail.
Keep humans at consequential boundaries
- The prepared CI-failure demo requires approval before creating the repair issue.
- Medin explicitly moved away from a fully unattended “dark factory.”
- Caveat: sponsored Port walkthrough; the two-day application and 20 pull requests are Medin’s demonstration, not an independent productivity study.
I encourage you to keep it simple and stick to the core loop.