Meta Excludes AI-Generated Code Volume from Engineer Evaluations, Returning to Essential Development Outcomes

News Overview

Meta has revised its internal performance evaluation criteria for engineers. The company has decided to remove "the volume of code generated using AI tools," which had previously served as a metric for some evaluations, from its HR performance indicators.

Background and Challenges

With the widespread adoption of AI, a practice known as "tokenmaxxing" emerged among engineers, where they sought to maximize the amount of code generated using tools like GitHub Copilot. This was aimed merely at inflating the output count produced by AI rather than focusing on fundamental problem-solving or product quality improvement. This phenomenon made it difficult to objectively measure productivity in performance reviews, serving as the backdrop for this policy revision.

Meta's Future Evaluation Policy

Moving forward, Meta is abolishing simple metrics that rely solely on the output results of generative AI. The company is returning to an evaluation system that emphasizes rigorous human code reviews, contributions to entire projects, foundational engineering skill sets, and the ability to handle technical complexities. This demonstrates Meta's stance that AI is ultimately a tool to assist productivity, and its usage should not distort performance evaluations.