Meta Engineer Burned $500K a Month on AI Tokens: Employees With Low Usage Faced Layoff Risk

Meta has laid off thousands of workers as part of its plan to transition to artificial intelligence (AI). The company reportedly relied on several standards to select employees for layoffs, including performance evaluations, department restructuring, and AI initiatives, with internal experimentation around AI usage metrics reportedly influencing engineering output assessments.

Part of that process included the short-lived leaderboard called Claudeonomics, allegedly created by an employee. The internal leaderboard, which reportedly ran from March, was also known as tokenmaxxing—a system that encouraged engineers to consume as many AI tokens as possible.

AI token consumption was important for an employee since it was a gauge to rank them and determine their efficiency. In internal accounts circulating among employees, higher token usage was informally interpreted as a proxy for AI-driven productivity, while lower usage raised concerns about adaptation to AI tools. That would in turn be essential in identifying which workers were worth keeping or letting go, depending on how they could cope with AI.

While it appeared ingenious, it was also costing Meta a fortune. According to an X user with the handle @sheiyuo, one Meta engineer burned roughly $500K per month in token consumption (roughly 300 million tokens in a month).

Glitch Found for Survival

Aware that ranking high on the board was essential for them to keep their jobs, some workers found a loophole to make sure they would be among the leaders. Some allegedly used idle AI agents that ran all day to make sure that they would be in the safe zone, artificially inflating their token usage to improve their ranking.

However, that initiative was also costly for Meta. It increased computing costs, placing the company in a worse predicament. Aside from that, the system was leaked as well, adding more controversy.

The board was shut down on 8 April 2026, allegedly at the behest of an employee to prevent further leaks.

However, the question now is whether the leaderboard lasted only days or as long as a month.

Although Claudenomics is widely said to have started in April, @sheriyuo claims it ran as early as March. If these claims are true, it suggests the leaderboard may have collected a large amount of data while experimenting with AI and identifying who could adapt to the new technology. Token usage patterns reportedly emerged as an informal signal in internal discussions about productivity under AI-assisted workflows.

Considering it cost Meta a fortune, it cannot be discounted that whatever data was collected in that short span would be used in future coding or AI models. Indirectly, it could be a means to determine which employees were worth keeping or not, based in part on observed AI tool engagement levels.

The key performance indicator (KPI) from using Claudenomics was arguably AI token usage, with internal chatter suggesting it was treated as a rough proxy for AI productivity during the experiment. Whether this would be the new normal that Meta would use internally moving forward is anyone's guess.

The only ones who can likely answer that are the remaining Meta employees. Roughly 7,000 were kept, reassigned to train AI models. While Claudenomics was reportedly shut down permanently, it could be revived again somewhere down the road to entice healthy competition once more, but with necessary revisions to avoid existing employees from cheating their way into being retained.

New KPI, New Cuts Looming

Recall that in a Reuters report, Meta was planning to cut potentially more than 10,000 employees this year. So far, they have already laid off about 8,000 employees, roughly 10% of its global workforce. Assuming they stick to that target of 10,000 layoffs, a potential benchmark for determining who deserves to stay could be their AI knowledge, including how effectively they use AI tools in daily engineering workflows.

With Meta bent on investing heavily in AI, the company has set between $115 billion and $135 billion in capital expenditure. The total expenses for the year are projected to reach as much as $169 billion, as the company continues its push towards AI systems, believing they can surpass human intelligence.