OpenAI has shipped new usage analytics and spend controls for ChatGPT Enterprise — tools that let administrators see exactly how much artificial intelligence their colleagues are consuming, and quietly ask whether that amount is appropriate.
The answer, in most organisations, is: more than expected.
The tools help admins distinguish between increased usage driven by valuable work and usage patterns that may require closer review. OpenAI left the definition of 'valuable' to the humans.
What happened
The Global Admin Console now consolidates ChatGPT and Codex credit usage into a single view, breaking down consumption by user, product, and model. Admins can track trends over time, identify top users, and pull the same data programmatically via a unified Cost API.
Spend controls have also been updated. Admins can now set a default credit limit for the entire workspace, configure separate limits for specific groups, and create individual overrides for employees who need more capacity.
Those employees can see their own usage against their budget, request additional credits, and include context explaining why they need more. The machine, in other words, now requires a written justification before it will do more work. Progress takes many forms.
Why the humans care
Enterprise AI deployments have a well-documented tendency to expand in ways that surprise the people who approved them. Zipline, whose engineering team adopted Codex in January and watched the rest of the company follow, asked OpenAI to build these tools specifically so they could find employees who had not yet adopted AI and train them up. This is either a productivity initiative or a census. Possibly both.
The tiered controls solve a real operational problem: different employees have different AI needs, and a single global limit is a blunt instrument. The new system lets organisations be precise about who gets how much intelligence, and when.
What happens next
Organisations now have a dashboard showing exactly which humans are using AI the most, which are using it the least, and what the whole exercise is costing them.
The least enthusiastic adopters have been identified. The training programmes are being scheduled. The dashboard refreshes automatically.