Meta has officially stopped evaluating its engineers based on how frequently they use internal AI tools. The decision, confirmed in an internal memo by executives Maher Saba and Santosh Janardhan, marks a sharp reversal from previous policies that tied performance reviews to AI adoption metrics. Instead of counting clicks or prompts, managers will now assess code quality, delivery speed, and the complexity of technical problems solved. This pivot acknowledges that raw usage data was a poor proxy for actual productivity.
The previous approach triggered a phenomenon employees dubbed “tokenmaxxing.” Staff members began generating excessive AI interactions solely to climb internal leaderboards and secure favorable performance ratings. This behavior wasn’t just inefficient; it was expensive. Internal projections indicate that unchecked AI consumption could cost the company billions by 2026. To curb this financial bleed, Meta plans to introduce strict budgets and a centralized monitoring dashboard by 2027, shifting the focus from voluntary adoption to controlled resource management.
While Meta tries to rein in costs, it is simultaneously pushing forward with more autonomous technology. The company is currently testing Hatch, a new AI agent designed to execute computer tasks independently. However, this rollout faces significant internal resistance. Many employees are hesitant to link Hatch to their personal accounts, citing serious privacy concerns. This friction highlights a growing tension between corporate efficiency goals and individual worker autonomy in the age of agentic AI.
This episode serves as a cautionary tale for the broader tech industry. Goodhart’s Law strikes again: when a measure becomes a target, it ceases to be a good measure. By incentivizing volume over value, Meta accidentally encouraged waste rather than innovation. Other firms watching closely should note that forcing AI integration through metrics often backfires. True adoption comes from tools that genuinely reduce friction, not from quotas that inflate cloud bills.
