Box CEO warns executives are falling for AI productivity hype

Box CEO Aaron Levie argued that chief executives are being misled by the sheen of AI demos, saying they are “uniquely prone to AI psychosis” because they sit too far from the “last mile” of work that makes outputs reliable in real operations. His warning landed in a debate that has grown sharper as executives pour money into AI while still struggling to turn it into measurable productivity.

Levie’s point was blunt: the hard part is not getting a model to generate an answer, a draft or a block of code. The hard part is everything that follows. He pointed to reviewing AI-generated code before deployment, checking AI-generated contracts against prior agreements, and folding outputs into existing systems without breaking workflows. In other words, the executive view often captures the “happy path” while missing the verification, governance and data-quality work that determines whether AI survives contact with production.

Levie said the smartest thing a CEO can do is use AI heavily, so leaders develop an appreciation for both the upside and the effort required to make it useful. That warning matters because the broader business debate has started to echo an older one about computing itself. Semafor linked the current moment to economist Robert Solow’s 1987 observation: “You can see the computer age everywhere but in the productivity statistics.” The comparison is apt because AI can look revolutionary in polished demonstrations while still needing major organizational adaptation before gains show up in the numbers.

The skepticism is not just philosophical. Deloitte’s 2025 survey of 1,854 executives across Europe and the Middle East found rising AI spending but elusive return on investment. MIT Project NANDA’s 2025 research went further, reporting that only about 5% of organizations were converting generative-AI pilots into meaningful operational or financial impact, with most pilots failing to reach production. That gap helps explain why many corporate plans sound more ambitious than the results they can prove.

Levie’s critique cuts to a larger risk for the economy: overconfidence at the top can distort hiring plans, spending priorities and investor expectations before the technology has earned that trust. The message from Box is not that AI is a mirage. It is that executives who mistake a convincing demo for durable productivity may be building strategy on the easiest part of the problem.