Summary
At OpenAI DevDay 2026, Danielle from OpenAI’s developer experience team interviews Itamar Friedman, CEO of Qodo, about the future of AI-driven code review.
Friedman describes Qodo as “Codex’s best friend”: it gathers an engineering organization’s tribal knowledge and feeds it to Codex, reviews Codex’s output as it codes, and acts as a governance layer that maps how software drifts over time. He distinguishes intelligence from “wisdom,” arguing that experience-based understanding of how a whole system behaves must be codified so coding and review agents can use it. He also traces the shift from prompt engineering to flow engineering to what he calls swarm engineering.
The conversation includes a customer example in which a change that looked fine locally risked a major outage because of downstream database effects, the kind of problem that is hard for both agents and human reviewers to spot. Friedman predicts that by 2027 reviewing findings on a pull request page will look outdated, with coding and review agents instead reaching agreement before involving a developer. A useful metric, he suggests, is how many times a developer must be prompted to complete a high-quality PR, and he explains how incidents can feed back so the same mistake is not repeated.
📺 Source: OpenAI · Published October 07, 2026
🏷️ Format: Interview
