Box founder Aaron Levie argued that many CEOs overestimate what AI agents can do because they are removed from the “last mile” of work that generates most value. Levie said executives prototype AI or generate outputs and then assume agents can do full tasks, while those who must verify and deploy AI—engineers and reviewers—face the detailed work of fixing bugs and catching hallucinations.
Levie’s Argument
Levie wrote on X that CEOs are prone to “AI psychosis” because they do not see the day-to-day work needed to make AI reliable. He advised CEOs to use AI extensively to learn both its upside and the real operational effort required.
Levie’s AI Position
Levie is not an AI skeptic; he posts optimistic commentary and invests in AI startups, and he has argued for “headless software” and agent-driven workflows. His view recommends hands-on experimentation rather than top-down declarations about automation.
Layoffs And Claims About AI
Tech sector layoffs have accelerated in 2026, with Layoffs.fyi reporting 115,430 job cuts at 152 companies through May, close to 2025’s total of 124,636 cuts across 275 firms. Many companies cite AI as a reason for reductions, and critics accuse some firms of using “AI” as a cover for other business decisions.
Examples From Industry
ClickUp CEO Zeb Evans said his company cut about 22% of staff after deploying some 3,000 internal AI agents, describing his goal as creating an organization where people run and review agents’ work. Evans framed the move as changing roles rather than cost-cutting, aiming for what he called a “100x org.”
Evidence On Productivity Gains
Academic reviews offer mixed evidence on AI’s productivity effects. A UC Berkeley meta-analysis found no robust relationship between AI adoption and aggregate productivity gains, while an NBER study reported perceived gains exceed measured gains. MIT researchers evaluating thousands of worker-agent tasks concluded agents are not yet delivering human-quality work broadly and estimated agents could reach a minimally sufficient 80–95% success rate on many text tasks by 2029.
Management Bottlenecks
Harvard Business Review research found that widespread AI use shifts bottlenecks to managers, who must authorize and coordinate increased output. Observers warn that if executives are unprepared for that coordination work, organizations risk chaos as AI tools scale internal output.
Featured image credits: SPARK Services
For more stories like it, click the +Follow button at the top of this page to follow us.