← Capa | Mercado

Meta drops token counts from employee performance reviews after AI leaderboard backfires

2026-11-04 · Fonte:InfoWorld

Meta drops token counts from employee performance reviews after AI leaderboard backfires

Meta executives say AI adoption dashboards and token counts will no longer factor into annual reviews, after internal leaderboards pushed engineers to game metrics rather than focus on core work.

Meta has stepped back from using AI usage metrics in annual performance reviews. According to a report by The Information, the company has told staff it "will not use AI adoption dashboards or token counts to evaluate impact." The reversal, communicated in an internal memo from executives Maher Saba and Santosh Janardhan, marks a retreat from earlier internal campaigns that encouraged heavy AI consumption — a practice internally known as token maxing.

In the memo, Saba and Janardhan directed managers to assess "output quality, velocity, problem complexity and scope taken on" rather than raw usage figures. The shift unwinds a culture in which Meta engineers competed to consume the most AI tokens, with their scores displayed on an internal leaderboard. The company found the contest was backfiring: employees were being diverted from their regular work, using AI to take on additional tasks purely to lift their scores.

Amazon ran into the same dynamic. The company implemented a leaderboard to track AI use and likewise found employees trying to game the system by using AI to complete unnecessary tasks; it has now deleted the leaderboard. Amazon does, however, monitor employees' AI usage for training purposes under a program introduced in April — but that information has not been used to measure employee performance.

The memo is not Meta's first sign of skepticism toward AI metrics. Earlier this year, Chief Technology Officer Andrew Bosworth told employees in a memo that "nobody should be using AI too…" The company's pivot underscores a lesson now echoing across the industry: aggressive adoption mandates, once framed as productivity levers, can distort behavior the moment they are tied to evaluation.

Tags: Meta · Amazon · Maher Saba · Santosh Janardhan · Andrew Bosworth · AI Adoption


Leituras relacionadas


萬安算交所 AIXX · AIXX · Exchange verificado · Ir ao mercado