Overview
- Last week Amazon shut an unofficial internal leaderboard called Kirorank after staff gamed it to inflate AI token counts and a senior vice president warned employees not to use AI just to boost usage.
- Other firms have shown similar moves: Microsoft closed internal leaderboards and restricted external coding agents, and Meta previously ended employee leaderboards after reports of extreme token consumption.
- Vendors and products are shifting pricing to charge by usage rather than flat seats, with GitHub Copilot moving to usage-based billing this week and companies like Anthropic and OpenAI reworking enterprise plans.
- Tokens are the units models process and heavy or purposeless token use raises direct API and infrastructure bills, a dynamic that reporting linked to very large internal consumption figures and sudden budget pressure at firms such as Uber.
- The near-term result is tighter governance: firms are rationing AI budgets, revoking or restricting tooling access, and refocusing rewards on measurable outcomes and model efficiency rather than raw token counts.