Box founder Aaron Levie said this week that tech CEOs are 'uniquely prone to AI psychosis,' arguing that executives championing AI rollouts are too removed from the actual work to judge whether the tools are delivering. The line landed against a measurable backdrop: DuckDuckGo said installs are up 30% after Google announced it would push more AI into its core search experience. Levie is not anti-AI — he is telling his peers to actually use the products before staking the company on them.
Levie's framing pins the problem on distance from execution. He said CEOs are sufficiently removed from the last mile of work that has to happen to generate value with AI, and that without touching that work they cannot accurately judge productivity claims. The comment circulated on social media before being unpacked on TechCrunch's Equity podcast.
“distant from the last mile of work that still has to happen to generate most value with AI”— Aaron Levie, Box founder
The DuckDuckGo number is the more concrete data point in the broader backlash story. A 30% lift in installs is small in absolute terms — DuckDuckGo remains a fraction of Google's scale — but the direction matters. It suggests a non-trivial cohort of users are actively migrating away rather than tolerating AI Overviews and conversational answers stitched into results.
Key facts
- 01Box founder Aaron Levie said tech CEOs are 'uniquely prone to AI psychosis' because they are distant from the actual end work.
- 02DuckDuckGo installs are up 30% following Google's announcement that it will push more AI into the core search experience.
- 03Google's AI returned that 'Google' has two P's when asked how many P's are in its own name.
- 04Long-time Google users, some with two to three decades on the product, still think of it primarily as an information retrieval system, not a shopping assistant.
Google's own execution has not helped. Asked how many P's are in 'Google,' the company's AI answered two. For users who have spent two to three decades treating Google as an information retrieval system, mistakes like that erode the one attribute the brand was built on.
The strategic tension is that Google's AI push is heavily oriented around commercial transactions — shopping, booking flights, tasks that end in a purchase — while the core user contract is about finding accurate information. Pushing one without protecting the other is what's driving defection to alternatives.
Google has tried to soften the rollout by keeping the older 10-blue-links experience accessible for users who want it. But the default direction is set, and the default is what most users will see.
A separate observation surfaced on the podcast: among frontier labs, there is a quiet convergence toward Anthropic's approach of narrowing what the product is for, rather than racing to bolt AI onto every surface. Google, by contrast, is still pushing in multiple directions at once and is vague about which use case it is optimizing for. That vagueness is part of why launches keep stumbling on basic factual tasks.
“there seems to just be this collapsing towards Anthropic's approach, this idea of really trying to understand what it is you want to offer people”— Sean O'Kane, TechCrunch journalist
On the workforce side, the AI-driven layoffs story is real and ongoing — ClickUp cutting 22% of staff for AI agents, which we covered earlier this month, is one of several examples. Levie's point is that the executives ordering those cuts are often the furthest from the code, the support tickets, or the customer calls that the agents are supposed to absorb. If the productivity slide looks clean but the end work is degrading, the CEO is the last person to find out.
There is a structural asymmetry here worth naming. Previous workplace technology shifts — cloud, SaaS, mobile — were largely bottom-up. Employees adopted tools they liked and IT eventually formalized them. The current AI wave is being pushed top-down by executives and the VCs funding their startups, both of whom are sold on the dream of a tiny team matching the output of a much larger one. Whether that dream survives contact with the actual work is the open question.
The counterweight is that plenty of workers do use these tools and find them valuable, and the AI labs have real revenue and real adoption to point to. Levie himself is not telling anyone to throw the tools out — he is telling CEOs to pick them up. Both things can be true: the tools work, and a layer of executives is overstating how well they work because they have not personally tested the claims.
For AI vendors, the practical signal in the DuckDuckGo number is that there is now a defensible market position in being the product that does not aggressively inject AI into the core flow. A year ago, even challenger search engines felt they had to ship AI features to stay credible. That pressure has eased. Expect more startups to test an explicitly restrained posture — AI available, but sandboxed from the primary experience — and expect the next round of enterprise software pitches to lean harder on measurable end-work outcomes rather than executive-deck efficiency claims.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
