Stop counting tokens, start counting what ships
A sponsored article from HPE argues that companies should stop rationing AI coding agents and measure results instead. The piece says a single agentic coding task can use 10 to 50 times more tokens than a basic chat prompt, which makes the most useful workflows the most expensive under per-token pricing.
The article cites several pressure points. It reports that Uber exhausted its entire 2026 AI budget by April after expanding its use of Claude Code. It says GitHub Copilot moved from flat-rate pricing to metered credits. It also cites a Gartner prediction that AI coding token costs will surpass average developer salaries by 2028.
The author argues that blanket usage caps measure consumption rather than output. Instead, it recommends tracking cost per merged pull request, defined as total AI-assisted development cost divided by the number of changes that pass review and are merged. It says rework and defect rates should be tracked alongside that figure.
The proposed approach has two parts. First, reserve frontier models for difficult planning and novel reasoning, and route routine work such as code generation, test creation, and refactoring to open-weight models the company operates itself. The article says benchmarks show some open-weight models can approach fr