Stop counting tokens, start counting what ships

The article argues that per-token pricing measures consumption rather than results, and proposes cost per merged pull request as a better metric for AI-assisted development. It recommends routing routine work to self-operated open-weight models while reserving frontier models for difficult tasks, and cites HPE's hybrid platform as cutting token costs 30 to 50% across more than 150 internal applications. It also reports that the top 5% of users generated most consumption, with the heaviest using roughly 450 times more than the lightest on the same license.