Post by DoorDash

Models out of the box have trouble reviewing DoorDash's codebase, even when we feed them the right context. Single-pass AI reviewers only caught 30% of the real issues in our PRs. Instead, we built a multi-model (Sonnet 4.6 + Opus 4.8) code review agent that caught 53.6% of issues at $3.91 a PR. Most enterprise code review solutions perform significantly below 50% recall on DoorDash PRs. To figure out which model and harness combinations to run, we built DashBench, our internal benchmark that compares multi-model code review systems against each other. It's how we've been able to experiment with and implement open source models, with Kimi K2.6 + Fable 5 currently leading as the most performant. Read the full DoorDash Engineering blog 👇