Machines of Code and Grace “I’ve often wished I could fork myself…”
8 snips
Mar 23, 2026 They explore defining AI feature owners to prevent accidental breakage. They use a card-shuffler example to show how overlooked interactions cause bugs. They explain making agents that own knowledge, invoke reviewers during planning, and update knowledge bases. They discuss dividing attention with sub-agents and documenting new feature owners to keep systems reliable.
AI Snips
Chapters
Transcript
Episode notes
Feature Interactions Are Invisible Dark Matter
- Feature interactions are the invisible, growing requirement that every new feature must not break existing behavior.
- Avdi illustrates with his MTG deck shuffler where two-faced cards repeatedly get missed when adding new card properties.
Assign A Knowledge Owner For Each Feature
- Do assign a feature owner who knows a feature's intent, implementation, and interactions before adding new features.
- Avdi created a feature owner (an actor) whose job is to review incoming plans and ensure two-faced cards remain supported.
Invoke Feature Owners During Planning
- Invoke feature owners as part of the planning step so they can flag interactions before implementation.
- Avdi implemented this by having an agent plan step that calls each owner with its prompt and knowledge directory.