Skip to main content

Meta Shifts AI Evaluation: Quality Over Token Counts, 93% Code Changes Now AI-Assisted

In a move that signals a maturing approach to AI integration, Meta has informed its engineering teams that AI token consumption will no longer be a metric in performance evaluations. This change, communicated through an internal memo, reflects a broader industry trend toward measuring the actual value AI brings, rather than the volume of usage.

The memo, issued by executives Maher Saba and Santosh Janardhan, outlines new performance expectations for all engineers. It explicitly states that Meta "will not use AI adoption dashboards or token usage to evaluate employee impact." This marks a departure from earlier practices where such metrics were considered.

Instead, the revised guidelines emphasize work quality and tangible outcomes. The company is steering its workforce toward a mindset where AI is a tool for enhancement, not a checkbox for compliance. The memo also reveals a striking statistic: 93% of code changes at Meta are now completed with the assistance of AI agents. This figure underscores how deeply AI has become embedded in the company's development processes.

But Meta is quick to clarify that the goal isn't to chase a high adoption rate. The real objective is to leverage AI to deliver higher-quality code and accelerate iteration cycles. In other words, it's about working smarter, not just faster.

This shift in evaluation criteria comes on the heels of earlier reports about Meta's internal AI management. In June, the company reportedly sent a memo to 6,000 employees through a platform called AI Gateway, designed to centralize AI usage oversight. Teams were given visibility into their own AI consumption, and the system would automatically flag any unusual spikes in token spending. At that time, Meta noted that internal AI usage was growing rapidly, with projected costs potentially reaching hundreds of billions of dollars this year.

Now, by decoupling performance reviews from token counts, Meta is acknowledging that measuring employee value through such metrics can be misleading. It's a recognition that the true worth of AI lies in the outcomes it enables, not the resources it consumes.

For engineers, this means a clearer focus on what matters: solving problems, improving products, and pushing boundaries. The pressure to maximize AI usage for its own sake is lifted, replaced by an emphasis on meaningful contributions.

As AI continues to reshape the tech landscape, Meta's approach offers a glimpse into how companies might balance innovation with practicality. It's a reminder that metrics should serve the mission, not the other way around.

Image

Key Points

  • Meta no longer uses AI token usage in performance evaluations for engineers.
  • The new guidelines prioritize work quality and actual impact over AI adoption metrics.
  • 93% of code changes at Meta are now assisted by AI agents.
  • The company aims for higher-quality delivery and faster iteration, not just AI adoption.
  • This follows earlier internal efforts to manage AI usage and costs through the AI Gateway platform.