Editorial illustration for Meta drops AI use from performance reviews after "tokenmaxxing
Meta Ends AI Metrics After Engineers Game Reviews
Meta drops AI use from performance reviews after "tokenmaxxing
Meta engineers spent months gaming their own performance reviews by burning through AI tokens for the sake of looking productive on internal leaderboards. Employees at the company called it "tokenmaxxing," a workaround born from Meta's decision to tie AI tool usage to how engineers get evaluated. The math didn't hold up. Chasing token counts doesn't mean better code, and the strategy left Meta with inflated AI costs and reviews that measured activity instead of output.
Now the company is reversing course. Executives Maher Saba and Santosh Janardhan laid out the change in an internal memo obtained by The Information, telling staff that AI dashboards and token counters are out as review metrics. Engineers will instead be judged on the quality, speed, and complexity of their actual work, a shift meant to strip out the incentive to farm token usage for appearances.
The stakes go beyond performance reviews. Meta's internal AI spending is on pace to hit billions of dollars in 2026, pushing the company toward budget caps and a centralized dashboard set to launch in 2027. That cost pressure sits alongside the memo's core message about what leadership actually wants engineers optimizing for.
Meta will no longer judge its engineers by how much they use AI tools. AI dashboards and token counters won't factor into performance reviews anymore, executives Maher Saba and Santosh Janardhan said in an internal memo seen by The Information. What counts now is the quality, speed, and complexity of the work.
Why this matters
Meta's reversal is a tidy case study in what happens when you tie compensation to a proxy metric instead of the outcome you actually want. Tell engineers that token counts earn them credit, and they'll generate tokens, whether or not the work improves. That's not a people problem, it's a measurement problem, and it should worry any founder or engineering lead currently building AI adoption into review cycles or OKRs. If you're rewarding usage, you're optimizing for usage, not for shipped quality or speed.
For builders inside AI tooling companies, there's a demand-side signal here too: Meta's internal AI spend was reportedly heading toward billions, partly inflated by gamed metrics. That's a preview of what "AI ROI" audits will start looking like across the industry as finance teams ask why token bills are climbing faster than output. Expect more companies to quietly walk back AI-usage mandates once someone runs the numbers. Watch whether Meta publishes anything on how "quality, speed, and complexity" get scored, because that's the harder problem they just inherited.
Related Reading
Browse all Meta news →
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Databricks finds multi-step agents beat single-turn RAG by 21% to 38% on STaRK
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
Related Topic OpenAI hires 630 ex-Meta staff as ChatGPT memory may turn data into ads
Business & Startups
Related Topic Meta AI Update Pulls From Your Calendar for Daily Briefings
LLMs & Generative AI
From Archives Axis Robotics Launches Browser-Based AXIS Engine With 207 Robot Tasks
From Archives OpenAI says AI "research interns" cut internal support requests by half
From Archives Meta's RPMs Rank ML Experiments Before GPU Run, Show 10% Gains
From Archives Meta's Audio AI Identifies 20+ Speakers in Real-Time Conversations
AI Tools & Apps
Common Questions Answered
What is 'tokenmaxxing' and how did Meta engineers use it to game performance reviews?
Tokenmaxxing was a workaround where Meta engineers artificially inflated their AI tool usage by burning through AI tokens to appear more productive on internal leaderboards. Since Meta had tied AI tool usage to performance evaluations, employees exploited this metric by generating high token counts regardless of whether their work actually improved, leading to inflated AI costs and misleading productivity measurements.
Why did Meta decide to remove AI usage metrics from engineer performance reviews?
Meta removed AI usage metrics after discovering that measuring engineers by token counts and AI dashboard activity didn't correlate with better code quality or actual productivity. The company realized that tying compensation to usage metrics incentivized employees to generate tokens rather than produce meaningful work, resulting in wasted resources and inaccurate performance assessments.
What new criteria will Meta use to evaluate engineer performance instead of AI token usage?
According to executives Maher Saba and Santosh Janardhan in an internal memo, Meta will now evaluate engineers based on the quality, speed, and complexity of their work rather than AI tool usage metrics. This shift focuses on actual outcomes and deliverables instead of measuring activity through dashboards and token counters.
What broader lesson does Meta's tokenmaxxing problem teach about using proxy metrics in performance evaluations?
Meta's experience demonstrates that tying compensation to proxy metrics instead of desired outcomes creates perverse incentives where employees optimize for the measured activity rather than the actual goal. This case study warns founders and engineering leaders that rewarding usage metrics will result in optimized usage, not improved performance, making it crucial to measure outcomes rather than activity in AI adoption initiatives and OKRs.
Further Reading
- Meta Pushes Its New AI Agent on Employees—but Eases Off on Tokenmaxxing - WIRED
- Months after Meta made AI tools mandatory for employees, company execs tell engineers performance reviews will not count AI use - Times of India
- Meta minimizes role of token maxing in employee evaluations - CIO
- Tech Workers Maxed Out Their A.I. Use. Now They're... - The New York Times
- The Pulse: 'Tokenmaxxing' as a weird new trend - The Pragmatic Engineer
