You Are Rationing AI By Job Title

OpenAI has just split the price of enterprise AI in two: a cheap seat, and an expensive one on top. Somewhere in your organisation, a spreadsheet is now being built that decides who gets which. That spreadsheet is a capability decision wearing a budget's clothes, and I don't think anyone in the room is calling it that.

The seat and the fuel

On 11 August, CIO reported that OpenAI introduced a Premium tier for ChatGPT Business, so enterprises can mix seat types across teams. Standard seats cost $25 a month, $20 on an annual plan. Premium costs $125 a month, $100 annually. Premium buys "5x more usage than Standard" and removes the five-hour usage cap. Standard keeps the limits. Sanchit Vir Gogia of Greyhound Research put the new logic in one line: "The seat is the engine an enterprise licenses. The fuel is now billed separately." Bhupendra Chopra of Kanerika was even plainer: "There's a predictable per-seat charge for everyday chat, and a separate metered charge for heavy agentic work."

So who gets the fuel? The next day, Fortune reported how CIOs and CTOs are already answering. At Samsara, CIO Stephen Franchetti has usage caps for non-technical employees while R&D got expanded access. He frames the caps as autonomy: it "puts people in the position where they're kind of in control and they can make choices as to which models they use." At Compass, CTO Shay Artzi: "We also put budgets for every engineer, so they are aware of how they're spending." At Docusign, CTO Sagnik Nandy narrowed context defaults and cut token usage by about half.

Read those three again and notice who holds the pen. Engineers get budgets and awareness. Non-technical employees get caps. Full access is following proximity to the tooling budget, because the people setting the line sit next to the people spending it.

The people you are capping

Here is why that should bother you. The people on the capped side, the case handlers, the policy staff, the people in finance and HR operations, are the ones who hold the knowledge of how the work actually runs. Not the process diagram. The work. The exceptions, the workarounds, the reason step four exists. Any serious redesign of a workflow runs through them or it fails.

And there is fresh evidence about what happens to exactly these people. On 6 August, Harvard Business Review published a two-year field study by Arvind Karunakaran of Stanford, Katherine Kellogg of MIT Sloan and Batia Wiesenfeld of NYU Stern, following generative AI innovation at two organisations. Their finding: what kept domain experts in the innovation process was scaffolds, structures supporting collective experimentation. Without them, the people closest to the work quietly drop out.

Quietly is the word that matters. Nobody resigns from an AI project. They just stop showing up to it, and the project carries on without the one thing that made it worth doing. Now add a usage cap on top. You have an organisation where the domain experts were already drifting out for lack of support, and the new pricing hands them a formal reason to stay out.

The meter is fine, the line is not

I want to be fair to the CIOs here, because the costs are real. Agentic workloads burn money in a way chat never did, and Nandy halving Docusign's token spend is just good housekeeping. Metering heavy use is a reasonable response to a real bill. The problem is not the meter. The problem is that the cap line is being drawn by who sits near the tooling budget, when it should be drawn by who the redesign depends on.

And to be honest in the other direction too: a premium seat on its own changes nothing. The HBR study is about structures, not access. An unmetered domain expert with no time, no peers experimenting alongside them and no route from their findings into the workflow will drop out just the same. Access is necessary. It is not sufficient.

One thing to do this week, and it needs no budget line. Take one workflow you genuinely want redesigned this year. On paper, write two names: who currently has unmetered access, and who holds the knowledge of how that work actually runs. If those are different people, and I suspect they are, move one premium seat across for a month. That is one line on an invoice and one email to IT. Then watch what the second person does with it, because that is your capability strategy showing itself.