Stop Counting Tokens, Start Counting What Ships
MIT Technology Review, Monday, November 2nd, 2026
An MIT Technology Review piece argues that AI coding costs should be measured per merged pull request rather than per token.
This MIT Technology Review piece says agentic coding tasks can use 10 to 50 times more tokens than a basic chat prompt, so the most valuable workflows can become the most expensive under per-token pricing.
It cites reports that Uber exhausted its 2026 AI budget by April after expanding Claude Code use, GitHub Copilot's shift to metered credits, and a Gartner prediction that AI coding token costs will exceed average developer salaries by 2028.
Rather than blanket usage caps, the author proposes tracking cost per merged pull request alongside rework and defects.
Costs can be lowered by reserving frontier models for hard planning and routing routine code generation and tests to cheaper models.