Meta Removes AI Token Consumption Metrics from Developer Performance Reviews
Meta executives Maher Saba and Santosh Janardhan issued updated engineering performance guidelines stating that the company will no longer use AI adoption dashboards or token usage to evaluate employee impact. The revised guidelines remove mandatory AI usage mandates for specific tasks, refocusing performance evaluations on code quality and actual business results. This decision marks an important shift in how major tech enterprises manage AI adoption, moving away from vanity metrics like token consumption toward meaningful software output. It provides a blueprint for tech companies working to accurately measure the return on investment and real productivity gains of AI development tools. Meta disclosed that 93% of internal code changes are already assisted by AI agents, demonstrating that high baseline adoption has already been achieved. This policy adjustment comes despite Meta managing internal AI infrastructure via an platform named 'AI Gateway' to monitor usage and alert teams when token spending spikes, as internal AI tooling costs could reach tens of billions of dollars.
## BACKGROUND
In large language models (LLMs), tokens are basic units of text—such as words or subword chunks—processed by the model, making token count a primary unit for measuring computational workload and infrastructure cost. To govern multi-model enterprise usage, organizations frequently deploy an AI Gateway, a centralized architecture layer that handles routing, rate limiting, and observability for LLM requests.