How we review 4x more AI code without reading every line
Ankit Jain · 7 November 2026
Read the originalAviator's nine engineers stopped reading AI-written code line by line and now review intent, evidence and open decisions, merging four times as many pull requests.
Ankit Jain is co-founder and chief executive of Aviator, a company that sells tools for shipping AI-written code. In a sponsored post on The New Stack, Jain describes how the company's own engineering team changed code review. After AI agents began writing most of the code, weekly merged pull requests rose from about 100 to more than 400. An AI reviewer, a four-hour review deadline and a self-review tool all failed, so the team rebuilt review around intent, rules, evidence and decisions.
- The team sorted a year of review history, 11,400 comments, into 212 candidate rules and 64 approved ones. Each rule has a scope, a test method, a source and a responsible team.
- A review agent running on a different AI model argues with the author's agent before the pull request opens and records every disagreement.
- An agent proves each acceptance criterion in a preview environment with screenshots, requests and logs, and each pull request lists every decision and who settled it.
- With the same nine engineers, merged pull requests rose fourfold, median time to merge fell from 2.1 days to 3.4 hours and reverts fell by 40 per cent.
An engineering team can sort its past review comments into rules a machine can check and decisions a person must make, and keep human review for the decisions. This is a vendor's sponsored account of its own team, and the figures are the company's own.
List every decision made during an AI-written change in its pull request, with where the decision came from and who settled it.
Sort a year of code review comments into repeated rules and one-off decisions, and turn the repeated rules into checks that run on every pull request.
Derived by Working Surface from the article; more in the Playbook. Source line: Decisions need a person, and a diff is the worst place to make them.
8 November 2026: teams using AI agents are moving human review from reading everything an agent produced to settling the decisions it hands back, with evidence attached.