AI makes it cheap to produce issues and pull requests at volume, and a tracked system like GitHub makes that volume visible. Hiten Shah's measure for whether any single contribution is actually useful: does it remove uncertainty for the next person, or does it create more work while looking like work. "An agent-written pull request that saves one person ten minutes and costs someone else an hour is debt with a diff attached." A sloppy issue still wastes a maintainer's time even though it took the submitter seconds to file. The failure mode is specific: the contribution transfers the cost of resolving ambiguity onto whoever has to review or maintain it, rather than absorbing that cost upfront. The corollary for someone who cannot read code but wants to contribute: the bar is concrete rather than technical. Bring context the maintainer does not have. Describe behavior precisely. Name the failure mode. Show the example. Make the next decision easier. That is software work because it changes the quality of a decision, even though no code was written. **Cross-Domain Connections**: [[Non Engineers as Agentic AI Orchestrators]] covers the adjacent pattern of domain experts directing agents directly; this note covers the specific quality bar for whether that direction helps or burdens whoever inherits the output. [[Two-Bucket Model for AI vs Human Code — API Boundaries Over Review]] proposes governing AI code by contracts instead of inspection; this note's uncertainty-transfer test is a way to judge whether a given contribution earned that trust in the first place. ## Source - [[I Don’t Read Code. That’s Why I’m Learning GitHub.]] — Hiten Shah (@hnshah), X/Twitter, 2026-06-16 — https://x.com/hnshah/status/2066761276945211442/?s=12&rw_tt_thread=True