Stop counting tokens, start counting what ships
As AI agents plan, test, and iterate through coding tasks, they can use 10-50x more tokens than basic prompts. Companies like Uber and GitHub are already facing budget strain and shifting pricing models. The piece argues that measuring cost-per-token rather than cost-per-shipped-feature encourages the wrong optimizations, and blanket usage caps may restrict valuable workflows while hiding waste from unproductive retries.
Key points
- AI coding agents consume 10-50x more tokens than simple prompts due to planning, testing, and iteration cycles
- Token-based pricing creates budget pressures visible at Uber and GitHub; Gartner predicts AI coding costs will exceed developer salaries by 2028
- Cost per merged pull request is a better metric than per-token consumption to distinguish valuable work from waste
Summary and key points written by AI from the headline and a short excerpt, so they can be incomplete or wrong. MIT Technology Review did the reporting; read the original for the full story.
Read the full story at MIT Technology Review
