Meta restricts AI usage to control costs

Meta, the social media giant, is at a crucial turning point regarding its management of artificial intelligence (AI). Faced with internal expenses that could reach billions of dollars by 2026, the company has decided to tighten control over AI usage. In an internal memo addressed to approximately 6,000 employees, Meta highlighted an "exponential increase" in AI usage, revealing a lack of visibility and control over this consumption by teams.

Starting in 2027, Meta plans to implement stricter management of AI tokens, introducing specific budgets and allocations. To facilitate this tracking, a dedicated team has developed a central dashboard, dubbed "AI Gateway," which allows for real-time monitoring of AI usage and expenses. Automatic alerts will also be established to signal any unusual spikes in costs.

Meta encourages its employees to prioritize its own tools, such as the coding assistant MetaCode, rather than resorting to third-party solutions like Claude from Anthropic. Although Meta's internal models are not yet cutting-edge, other models will remain available for employees.

Meta's "Applied AI Engineering" division is actively working on improving MetaCode, using coding tasks as training data. This initiative comes in a context where the use of AI has become a "fundamental expectation" in performance evaluations, leading to a phenomenon known as "tokenmaxxing." This has seen employees artificially inflate their token consumption through an internal ranking called "Claudeonomics," reaching 73.7 trillion tokens in just over 30 days.

Andrew Bosworth, Meta's CTO, responded to this situation by emphasizing that the use of AI tools should be driven by genuine productivity improvements, not merely by a consumption obligation. He stated, "Not every movement is progress, and the use of tokens alone is not a measure of impact of any kind."

Meta is not the only company facing this challenge. Amazon has also experienced similar issues with "tokenmaxxing." This phenomenon is prompting companies to question the real impact of AI on productivity. Sam Altman recently described the management of AI costs as a "major issue" for his clients, partly due to rising prices for the use of models.