Meta is overhauling its internal AI governance after discovering that employees racked up billions of dollars in unchecked usage costs. An internal memo sent to approximately 6,000 employees warned that unmanaged AI consumption is on track to cost the company billions by 2026, according to The Information.
Previously, individual workers and teams had no visibility into their own AI consumption. This lack of oversight created a culture of tokenmaxxing, where employees artificially inflated their usage to meet performance review expectations. One internal leaderboard, dubbed "Claudeonomics," saw users generate 73.7 trillion tokens in just over 30 days.
Reining in internal AI costs
To reverse this trend, Meta will implement strict budgets and allocations for AI tokens starting in 2027. A dedicated engineering team built a central AI Gateway dashboard to track spending and usage in a single location. The company also plans to deploy automatic alerts to flag unusual cost spikes before they escalate.
Managing these expenses has become a central focus of corporate strategy, as organizations must now balance rapid tool adoption with strict financial oversight. Meta is actively steering employees away from third-party models like Anthropic's Claude and toward its proprietary coding assistant, MetaCode. Alternative models will remain accessible, however, since Meta's current offerings do not yet match frontier capabilities.
Productivity over volume
Meta's CTO Andrew Bosworth addressed the issue directly in a separate memo, criticizing the focus on raw volume. "Nobody should be using AI tools just for the sake of using them. All motion is not progress and token usage alone is not a measure of impact of any kind," Bosworth said. He emphasized that these systems should only be deployed when they "genuinely allow us to do better work, faster."
Engineers within the new Applied AI Engineering division are now tasked with improving MetaCode. They are generating specific coding tasks to serve as high-quality training data, moving away from the previous metric of raw token generation.
Why this matters for managers
The Meta situation highlights a critical risk for technology leaders: incentivizing AI adoption without guardrails invites financial waste. Effective AI for Management requires tying system usage to specific business outcomes rather than raw consumption metrics. Establishing baseline usage budgets and monitoring dashboards early will prevent internal tooling from becoming an uncontrolled expense.
Your membership also unlocks:
