Working with AI: sort your work by AI-verifiability, then design verifiers or be the judge

danshipper · x · 2026-10-18

@leeknowlton shares a framework for productive AI use, illustrated by a decision-tree flowchart on whether your work is AI-verifiable:

Split work into AI-verifiable vs. not — the test is whether AI can independently judge if the completed work is good.
Verifiable: build a verifier (or system of verifiers), then let AI run at full throttle; you're no longer the bottleneck, so scaling tasks costs no extra attention.
Maybe: build an AI judge, but validate it with evals (he points to @hammermt for how). A good enough judge turns work into verifiable; some judges are trivial to build, others not worth it.
Not verifiable: work the old way with AI help — find and fix bottlenecks, do one or two tasks at a time, avoid context switching, protect your focus.

More from coding & agent