DoorDash Raises AI Code-Review Recall to 65.2% by Splitting Scans Across Scout and Verifier Models

aiai-productsai-agents-codingai-modelingai-research-evalsai-enterprise-tools 3 posts · 3 accounts

DoorDash deployed a two-stage AI code review pipeline that splits pull requests between a scout model and a verifier, raising weighted recall to 65.2% and lowering costs to $3.81 per pull request. The strongest benchmark run paired Kimi K2.6 as the scout with Claude Fable 5 as the reviewer, while a baseline single-pass system previously captured only 30.7% of weighted issues.

The configuration achieved a 75.3% F1 score, surpassing single-pass reviews but trailing a Composer 2.5 and GPT 5.5 medium pairing that recorded 92.2% precision with significantly lower recall. The results show that routing verification tasks to dedicated models optimizes production cost and defect detection.

From the sources (3 posts)

@rohanpaul_ai

DoorDash showed open-source models can lift AI code review when used together. And that harnesses now matter. Single-pass AI reviewers caught only 30.7% of weighted PR issues. The company’s production reviewer raised that to 53.6%, while

@pmarca

RT @andyfang: With our internal coding benchmark, we're able to confidently introduce open-weight models into our AI code reviewer w/o degr…

@sriramk

very impressive use of open weight models and using both frontier tokens but also passing off lower-cognition work to other models.

Ollama Raises $65 Million in Funding as Open AI Tool Tops 9 Million Builders Lovable Raises $300 Million in Menlo Ventures-Led Round at $13.2 Billion Valuation