Echoing throughout New York Tech Week this year were conversations from business leaders about AI’s ROI, as companies begin cutting back on token spending.

CEOs are trying to prove to their boards that the cost is worthwhile during a time of “massive over-investment” industry-wide, IBM Vice Chairman Gary Cohn said Thursday. The world will need AI, but “the question is, are we going to need everything we’re building?” In an era of tokenmaxxing, Cohn said that so far, the ROI has “not been nearly as high as people might think.”

Others sang a different tune. Two startup CEOs told Semafor that companies are reducing their AI spending not because the tools aren’t useful, but because the results are too difficult to measure.

Most of the costs for companies come from software engineering workflows, with the market for coding capabilities jumping sevenfold from 2024 to 2025, to $4 billion, according to Menlo Ventures. But it’s long been difficult to determine the ROI of software engineering.

In the near term, the AI companies that will benefit are the ones that automate work that is measurable, like sales and customer experience, the two startup CEOs said. In the meantime, the best way for engineers to be free of the token leash is for companies to figure out how to measure the value of their tasks.