Cheap AI tokens are running out

James Stanier, chief technology officer of Nordhealth, tells engineering leaders to make AI usage visible and match AI models to tasks before per-token billing raises costs about tenfold.

James Stanier, chief technology officer of the software company Nordhealth, writes that the flat-rate AI plans many engineering teams use are subsidised and may end. His own team is on such a plan, but he estimates the same usage would cost about ten times as much at per-token prices. He sets out habits a team can build now so that a metered bill does not arrive as a shock.

• Engineers frustrated by company usage limits often switch to personal plans, which sends company code through accounts the company does not control.
• Every engineer should know how to see their own usage and limits and which AI model suits which kind of task, with a shared chat channel for examples.
• Managers see their own team's usage, while engineers see department-wide figures without names; Stanier warns that ranking people by usage leads them to use AI for its own sake.
• He suggests using the most expensive AI models for large exploratory tasks that produce plans, then giving the plans to smaller models. He notes that the effort setting changes cost as much as model size.

An engineering manager can start sharing usage figures and model-choice guidance while costs are still predictable. The article was metered on the publisher's site and was read in full.

Derived by Working Surface from the article; more in the Playbook. Source line: One thing to avoid is ranking people by tokens, because you get tokenmaxxing, usage chased for its own sake, which is the opposite of what you want.

30 October 2026: as AI agents do more of the building, teams decide on purpose which work people keep and set up tests, usage figures and baselines before handing over more.