Stop counting tokens, start counting what ships
As agentic coding workflows drive up token consumption and API bills, enterprises are exploring hybrid hosting and model-routing strategies to control inference costs.
Reported by MIT Technology Review ·
What happened
In a sponsored post, HPE argues that enterprises using AI coding agents should evaluate spending by the cost per merged pull request rather than raw token consumption. The piece advocates for a hybrid infrastructure approach that routes routine coding tasks to on-premises open-weight models while reserving expensive frontier APIs for complex planning.
Who it’s about
- HPE
- Uber
- GitHub Copilot
- Claude Code
- Gartner
Source
MIT Technology Review
Stop counting tokens, start counting what ships