CEOs Who Think AI Replaces Their Employees Are Just Bad CEOs

In a Techdirt piece that's trending on HN (360 points, 143 comments), Mike Masnick and Box CEO Aaron Levie dissect a pattern: CEOs, disconnected from day-to-day execution, see impressive AI demos and suddenly mandate tool adoption across the company. The result? Counterproductive token wars, resentment, and poor outcomes.

Levie's diagnosis: CEOs suffer from what he calls "AI psychosis" — they see the happy path (a working prototype, a generated contract) but miss the 10-20 additional steps needed for production-grade results. A CEO might demo Claude Code generating a product in minutes, but they didn't have to review the code for security, fix bugs in CI, or wire up past contracts for legal compliance.

Why Forcing AI Fails

• Token leaderboards are the worst idea. Masnick calls them "the dumbest way to encourage LLM usage." Good usage treats tokens as scarce resources; counting volume incentivizes waste and counterproductive queries.
• Mandatory use kills adoption. No one forced to use tools learns them well. Willing, self-motivated exploration is the only path to mastery.
• CEOs are "sufficiently distant from the last mile of work." They see a prototype and assume that replaces entire engineering, legal, and compliance teams — ignoring the details that make products safe and scalable.

Agentic Coding Tools in the Crosshairs

Levie specifically targets agentic coding assistants like Claude Code. The gap between "I built a thing" and "anyone can build a thing well at scale" is massive. A demo might work in isolation, but production demands security audits, accessibility checks, legal reviews, and integration with existing systems — work that can't be shortcut by an agent.

Masnick and Levie agree: the best CEOs use AI heavily themselves to learn its real limitations, then emerge with balanced expectations. The article is a must-read for any developer dealing with top-down AI mandates.