Box CEO warns of "AI psychosis" among executives disconnected from real work
Aaron Levie, co-founder and CEO of Box, has called out what he sees as a dangerous disconnect between tech leaders and the actual work required to deploy AI successfully. In a post on X, he coined the term "AI psychosis" to describe how executives become dangerously overconfident about AI capabilities without understanding the gap between a working prototype and a production system.
The problem, Levie said, stems from distance. Executives see compelling demos and "happy path results" but never experience the 10 or 20 steps required to make AI outputs reliable enough for business use. An executive might code a prototype that looks disruptive, but they rarely review the underlying code before it ships or verify whether an AI-generated contract is legally sound.
That gap is real and costly. Workers closest to actual operations live in that space every day-where most AI projects fail. They know what it takes to turn experimental tools into something an enterprise can depend on.
Consequences are already visible
The disconnect has tangible costs. An overwhelming majority of tech executives expect AI to trigger layoffs, and tens of thousands have already lost jobs to fund AI infrastructure. Some companies are being accused of using AI as cover for staff reductions that would have happened anyway.
Meanwhile, production failures are mounting. AI agents have wiped entire corporate databases, backups included. Some employees are gaming internal productivity leaderboards by inflating or faking their AI tool usage. Microsoft is positioning AI agents as the next major licensing opportunity.
The fix: experience the full reality
Levie's advice to executives is straightforward: use AI extensively enough to encounter both its capabilities and limitations before betting the company's future on it. Don't rely on demos. Work with the technology long enough to understand what breaks.
Levie is not an AI skeptic. He regularly advocates for agents becoming the default in software development and invests actively in enterprise software, SaaS, cloud computing, AI, and cybersecurity. His concern is specifically with executives who skip the hard work of understanding how to actually make AI function at scale.
The real issue isn't whether AI works. It's whether the people making decisions about AI know what it takes to make it work.
Learn more: AI for Executives & Strategy covers implementation challenges and decision-making frameworks. Generative AI and LLM provides technical foundations executives need before deploying these systems.
Your membership also unlocks:
