The AI Cost Question Every Leader Should Be Asking

Hanah-Marie Darley, Co-Founder and Chief AI Officer, Geordie.

The invoice is only the beginning.

Spend is the natural starting point for any conversation about AI cost. What I’ve found to be the more useful question is what that spend enabled: What was the agent trying to achieve, and was the outcome worth the resources involved?

AI cost is moving from a technology line item to an operating question as agentic work becomes the new normal in the enterprise. That is a welcome development. As agents research, analyze, build and communicate across the enterprise, their work is beginning to shape how teams allocate time, resources and attention.

Most organizations can already see model invoices, token consumption and aggregate usage. Those measures answer an important accounting question, but they do not yet explain what the organization achieved from that spend.

A higher bill could reflect an agent completing a complex piece of analysis that would otherwise take a team days. It could also reflect a task that repeated itself, selected a more expensive model than necessary or called tools without moving the work forward. The number alone cannot tell an executive which is which.

This is why one of the most useful units of AI cost is work.

Work gives cost its context.

I find the unit of work offers a practical path from spend to a decision. It allows leaders to ask what an agent was trying to accomplish, what resources it used, how it approached the task and what happened as a result.

The model is only one part of that picture. The cost of an agent’s work is shaped by its behavior, including the decisions it makes, the tools and information it uses, the retries it requires and any human intervention along the way. Those elements matter because they explain the economics of the work, not because they are interesting technical details in isolation.​