MIT Technology Review: Shifting from Counting Tokens to Measuring Outputs in the AI Agent Era

MIT Technology Review reported on 2026-11-02 that as AI agents take on tasks requiring repeated loops, the industry is shifting how it measures a model's value, moving away from cost and toward what the model actually delivers. The report gives coding as an example: when a developer hands a task to an AI agent, it does not just answer once. It plans, generates code, runs tests, and loops back to correct errors. A single task can consume 10 to 50 times more tokens than a basic chat prompt, and the report notes that the more capable an agent becomes, the more calls it makes. The coverage describes a shift in perspective among industry observers rather than a specific announcement by any company. No vendor, product, or version is named, and no follow-up developments, concrete metrics, or timeline for adoption are provided. The account stops at the observation that teams are beginning to evaluate agents by the work they complete rather than by the tokens they spend, leaving the broader implications open.