Datadog, the observability giant, is using OpenAI’s coding agent Codex to level up its code review process. The goal? Catch system-level risks that human reviewers might miss, especially in complex distributed systems. Traditional AI review tools often act like fancy linters, flagging superficial style issues while ignoring deeper architectural impacts. But Datadog’s AI DevX team built a clever test: they replayed past incidents, feeding the original pull requests to Codex as if it were reviewing them fresh. In over 22% of cases, engineers confirmed Codex’s feedback would have prevented the incident. That’s huge for a company where reliability is everything. The key insight? Codex doesn’t replace human reviewers—it complements them by spotting blind spots that even senior engineers can overlook. Why it matters: Datadog’s results suggest AI-powered code review could save companies money, headaches, and customer trust by catching systemic bugs before they hit production.