AI capacity has a price, and that price moves. Not as a quote on a board, but as the sum of what it costs to have a task done by a model versus by an employee: compute costs, licenses, the time of human oversight still required, and the cost of correcting errors. That sum changes every quarter, sometimes every month, and each time it changes, the boundary shifts somewhere between what AI can take over, what still requires oversight, and what remains human work. A strategy built on that boundary quietly grows outdated if nobody tracks the price.
The price of AI capacity is not a single number but a stack of sources. Model vendors publish prices per processed unit, and these typically fall while quality rises, although the pace differs by provider and by type of task. Then there are the indirect costs: the time an employee spends reviewing and correcting AI output, which does not disappear as long as oversight remains required. And there is the comparative side: what it costs to have the same work done externally, with or without AI in the process. Anyone who looks only at the list price of an AI subscription sees a fraction of the picture.
A quarterly rhythm suits most sectors: often enough to spot a structural shift before a competitor has acted on it, not so often that you mistake noise for signal. In sectors where model vendors and applications succeed one another quickly, a monthly look may be needed at the tasks closest to the boundary. The frequency matters less than whether the look is tied to a decision; a signal nobody consults when making choices is not an instrument, it is an archive.
If the price of AI capacity for a given task drops below the cost of human execution, that task shifts from human work into the zone where oversight suffices, or from oversight to full takeover. That is a fact about the work, not about the people currently doing it. It means an assumption in your strategy — about the size of a team, the lead time of a process, or the cost level of a service, for example — has reached the point where it needs to be reviewed. It does not automatically mean something must happen; it means there is something to assess.
A falling price is not an instruction to reduce staff, and this signal is not intended as grounds for a dismissal decision. What an employer does with its staff falls under its own legal requirements and its own considerations; this is about facts about work, not personnel advice. A price change also does not mean a task's takeover follows automatically: between being possible and actually happening there is often still an implementation process, a quality standard to be met, or a customer who prefers to speak with a human. And it does not mean every part of a process shifts at the same pace — some steps remain human work for years while others tip within a quarter.
In some organizations this is already part of a fixed rhythm: a small team that, every quarter, compares the cost of AI execution against their own process hours, tied to the list of assumptions the strategy rests on. In other organizations that rhythm does not exist, usually not out of unwillingness but because nobody has been explicitly assigned ownership of the question. The strategy was calculated at the outset based on the then-current ratio between human and machine, and nobody has been tasked with continuing to track that ratio. The difference rarely lies in the sector; it lies in whether someone has been designated to look.
The price of AI capacity does not stand on its own. It moves together with what competitors without staff are already doing with the same technology, with the regulation that determines what AI is allowed to decide in your sector, and with what your own employees are already doing with AI themselves, independent of policy. A price drop that coincides with relaxed regulation or with entrants already operating without staff carries more weight than a price drop occurring in isolation. Anyone who looks at signals separately misses the moment when, together, they invalidate an assumption.
The price of AI capacity tells you what becomes possible in a general sense. It does not tell you which part of the work in your own company, task by task, actually falls within that boundary. That question — what can be taken over by AI here, today, task by task, what requires oversight, and what remains human work — is answered by FTE TO AI's work scan on a per-task basis, independent of the general trend. For the broader question of whether your strategy is already shifting on points like these, you can review how to recognize that a strategy has become outdated, and for setting up your own rhythm of signals there is an explanation of how to set up a horizon scan yourself.
You can start by naming the assumptions your current strategy rests on: which assumption about costs, capacity, or lead time would no longer hold if AI capacity became a third cheaper. A free assumption check offers a brief exercise for this: you name the key assumptions, and for each one it becomes visible when it was last confirmed. The full proofing process, in which this is placed alongside the external signals and a self-plot by management, is under construction.
Stel uw vraag. Vaak zit de echte vraag een laag dieper — daar mag ik naar vragen.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.