← back
How to Kill the Code Review — Ankit Jain, Aviator
Takeaway
Move review toward intent, behavioral evidence, and reusable checks while preserving the collaboration that line-by-line review previously supported.
Summary
- Jain separates code review into semantic correctness and team alignment, arguing that mentorship, shared understanding, and architectural discussion must survive automation.
- Static specifications miss decisions made during implementation; prompts and user responses contain valuable intent that is often discarded after a pull request is created.
- An AI slop registry turns recurring human review comments into reusable guardrails rather than requiring the same corrections repeatedly.
- The proposed verification loop converts session decisions into acceptance criteria, combines them with accumulated rules, and runs an end-to-end test plan in a preview environment to produce reviewable evidence.
code-reviewverificationacceptance-criteria
Original description
Over 30% of changes now merge with no review at all, and the wait on the ones that do get reviewed is four times what it used to be. Ankit Jain's read is that the debate about when we stop reading code line by line is already over, because we stopped. His sharper point is what replaced it: an AI writes the code, an AI reviews the code, the two go back and forth in a web UI, and a human skims the thread and merges. When AI reviews and nobody reads, he says, we have configured the wrong thing. He is also here to correct his own five layer trust model from a few months earlier, which missed that review was never only about correctness. It also carries knowledge sharing, mentorship, architectural feedback, and onboarding, and that half has to survive. Spec driven development does not rescue it, because a spec written up front with no feedback loop is the 1970 waterfall model, and the decisions that actually matter end up in the prompts, which teams throw away the moment the pull request opens. His proposal keeps them: capture the session, turn those decisions into acceptance criteria, pair that with a registry built from your own recurring review comments, and generate a test plan that a verification system runs against a live preview. The review surface becomes intent and evidence instead of the diff. Speaker info: https://x.com/ankitxg / ankitjaindce https://www.latent.space/p/reviews-dead Timestamps: 0:00 - The five layer trust model, and what it got wrong 1:17 - We already stopped reviewing, and AI reviews nobody reads 3:05 - What code review was actually for 4:21 - Spec driven development is waterfall again 5:53 - Intent lives in the prompts, which we throw away 6:42 - The AI slop registry 7:58 - Session to acceptance criteria to test plan 11:51 - Deterministic where you can, and reviewing intent not the diff 14:12 - Homework: mine your last 1,000 review comments