What happens when companies become too AI-pilled?

Box founder Aaron Levie called it "AI psychosis": the moment when executives, bedazzled by model benchmarks and cost-cutting fantasies, greenlight mass layoffs to replace roles they don't actually understand with AI systems they haven't tested at scale.

ClickUp cut 22% of its workforce recently, betting on AI agents to fill the gap. 2026 tech layoffs are already nearly matching all of 2025. The pattern is predictable: CEO reads a TechCrunch headline about GPT-5.5's coding prowess → CEO decides the entire support team can be replaced → CEO discovers, six months later, that what AI excels at on a benchmark doesn't map to what your actual customers need when they're angry at 2 a.m.

Levie's point cuts deeper: the people making these decisions—mostly non-technical executives—are the least equipped to assess what a job actually requires versus what an AI demo claims to do. A customer success manager who's been with your company for five years knows why customers churn. An LLM knows patterns in chat logs. These are not the same thing.

The fallout: reduced service quality, customer frustration, and (the delicious part) rehiring the people you just fired when the agents fail. It's the AI equivalent of "move fast and break things," except the things that break include your brand and your team's ability to ship anything that doesn't suck.