Automating Low-Risk Code Review at Meta: RADAR, Risk Calibration, and Review Efficiency

I'm Matt Wood, and this is For Your Information. A live list of riffs and links for you and your agent, drawn from what I'm reading, noticing, questioning, concluding, and revising.

The problem: At Meta, significant lines of code per human-landed diff grew by 105.9% year over year and per-developer diff volume rose 51%, with agentic AI responsible for over 80% of that growth. Meanwhile, the share of diffs receiving timely review has declined, exposing a widening gap between code supply and reviewer bandwidth.

arxiv.org →

Connections

Supports

How Frontier Teams Are Reinventing AI-Native Development Meta's data showing 105.9% YoY growth in lines-per-diff from AI coding provides concrete evidence for the productivity multipliers frontier teams report.

Related to

Formal Methods at Jane Street: Agentic Coding Changes the Calculus Both describe scaling agentic verification/review processes (Meta's RADAR risk calibration, Jane Street's proof assistance) to handle exploding volumes of AI-generated code.

Challenged by

OpenSSH 10.5 Release Notes OpenSSH's experience of being flooded with low-quality AI-generated security reports directly challenges Meta's RADAR approach of automating code review at scale — both reveal the signal-to-noise problem when AI generates security/review findings en masse

Type:	link
First published:	Saturday, 30 May 2026 at 12:00 UTC
Location:	https://arxiv.org/abs/2605.30208
URL:	https://mattwood.fyi/i/6a1cba95