The pitch and the price list
The pitch says everyone. The price list says premium. The Wall Street Journal reports that AI agents have such expensive computing needs that companies are offering them only to premium customers, or limiting who can access them. Its headline asks the right question: who shoulders the cost?
Rationing is a number in disguise. A chat reply is one model call. An agent finishing a task makes many, reads its own output, calls tools, tries again. When a company puts that behind its top tier or behind a usage cap, it is telling you the cost of a finished task is above what the standard subscription brings in. Nobody has to publish the margin for you to read it.
There are three places the bill can go: the customer, through tiers and caps; the provider, through losses carried as growth spending; or the product, through a cheaper model doing the work behind the same name. The Journal has read the headline and summary only, so which companies are doing which is not something we can list. The third is the one buyers rarely see.
The milestone
Watch for the first large provider to publish a price per completed agent task and hold it for two quarters. Until then, ask your vendor what the cap is, what happens when you hit it, and which model actually runs after that point. Everyone invited, some admitted, no rate card.
