newsfilter
← Interviews
AI Engineer28 min

60% of Our Data PRs Now Merge With No Human Review - Chetas Joshi, Robinhood

Chetas Joshi

Conference Presentation, Keynote, Talk

Robinhood implemented a hybrid PR review system to resolve the bottleneck where AI-generated code changes outpaced human safety verification, deploying deterministic scripts for factual checks alongside an AI agent that scans for unregistered risks. This architecture successfully automates approximately 60% of data pipeline reviews by escalating high-sensitivity scenarios to humans, resulting in a 75% reduction in manual effort and a 30% drop in incident rates as of late 2026. While the current setup runs check layers in parallel, the organization plans to evolve toward a unified classifier that grants the agent real-time access to production facts for more precise risk assessment.