Meta removes AI tool usage as a factor in engineer performance reviews after internal "tokenmaxxing" gaming behavior

According to The Decoder, citing The Information, Meta is removing AI tool usage metrics (dashboard activity, token counts) as a factor in engineer performance reviews. The change follows an internal memo, reportedly from executives Maher Saba and Santosh Janardhan, describing how the old metric encouraged "tokenmaxxing", engineers gaming internal leaderboards by racking up AI usage without producing real value. Reviews will instead weigh quality, speed and complexity of work. Meta says it plans usage budgets and a central AI-usage dashboard starting in 2027, though that is a stated intention, not a confirmed rollout.

This is a concrete reversal of an internal HR policy at one of the largest AI-spending companies in the world, useful as a real data point in the broader debate about how to measure AI-driven productivity. It matters most to enterprises and engineering leaders considering similar usage-based incentive schemes: the lesson is that raw usage metrics are easy to game and can inflate costs without reflecting real output. For Meta engineers specifically, this removes pressure to inflate AI usage for review purposes, but day-to-day impact outside Meta is limited since this is an internal process change, not a product or technical announcement.

There is no primary source here: this is The Decoder relaying The Information's account of an internal memo, with no link to the memo itself and no on-record Meta confirmation. The framing of AI costs "heading toward billions in 2026" is asserted, not sourced with figures, and should be treated as unverified context rather than a hard fact. The 2027 usage-budget and dashboard plans are forward-looking claims that may change or never materialise.

1. 01Whether Meta officially confirms this policy change in a public statement, blog post, or leaked internal document with verifiable detail
2. 02Whether other large tech employers (Google, Amazon, Microsoft) announce similar reversals of AI-usage-linked performance metrics over the next 6-12 months
3. 03Whether Meta's 2027 usage-budget and central dashboard plan actually materialises, and what specific metrics replace the old token-count approach