Developers at Meta, Microsoft, and Salesforce are deliberately burning tokens to inflate AI usage metrics. The practice, called tokenmaxxing, turns cost into a performance signal: hit the usage target, look productive, repeat. This is not a fringe behavior. It is happening at scale inside some of the largest engineering organizations on earth.
Simultaneously, Uber burned through its entire 2026 AI token budget in three months. Anthropic has already pulled enterprise plan subsidies. The math is breaking down fast, and per-engineer AI budgets are likely the next mandated control across the industry. The people gaming metrics now are accelerating that outcome.
The full piece also covers Cal.com moving core code to a closed repo, framed as an AI threat response but possibly just a long-overdue business model shift, plus Vercel open-sourcing its agent factory tooling and new AI usage guidelines landing in the Linux kernel. The tokenmaxxing section alone is worth your time because it shows how quickly incentive structures warp when adoption metrics replace output metrics.
[READ ORIGINAL →]