AIBRIEF
A WSJ report and small operator evidence point to the same shift: corporate buyers are choosing which model handles each task, and the priciest tier is starting to look like a specialty product.
Cursor's head of communications Mike Saeks told The Wall Street Journal that his company's customers now pick which AI model handles each task, and the most expensive option no longer wins by default. "The most powerful and expensive AI models aren't necessary for relatively mundane tasks," Saeks said. Cursor, the AI coding editor, sits in the middle of that shift because its product routes work to whichever model a user chooses.
A Hacker News thread on the WSJ piece pulled in a small-startup operator who described mixing Claude and OpenAI models across a capability spectrum, paying modest monthly fees on base plans for roughly 10x the output, with cost as one factor and capability fit as the other.
The WSJ piece frames the driver as "ballooning costs" and adds a wrinkle: some of the cheaper models now in the mix are built in China, sitting alongside US providers inside corporate stacks. The shift is real, but two things stay uncertain. Savings numbers so far are anecdotal rather than audited, and putting the wrong task on the cheap tier costs more than it saves. Tiered routing only works when the routing itself is right.