If you want the blunt version: “AI-pilled” companies risk becoming confidently wrong. The TechCrunch discussion around Box founder Aaron Levie’s warning is basically a case study in executive overreach—leaders with the most power over headcount are often the least familiar with the messy, tacit, judgment-heavy parts of the jobs they’re trying to replace.
<> That’s the core problem with a lot of AI strategy right now: it is not automation first, it is assumption first./>
Levie’s phrase for this kind of thinking—“AI psychosis”—lands because it captures the mismatch between hype and operations. It is easy to look at a workflow from 30,000 feet and conclude that an AI agent can do the job cheaper. It is much harder to understand the invisible labor underneath it: exception handling, coordination, context-switching, customer empathy, internal politics, and the thousand tiny decisions that keep a business functioning.
ClickUp is the example that makes the debate concrete. In the TechCrunch discussion, the company is described as cutting 22% of its workforce while planning to use 3,000 AI agents instead. The company was also described as serving 10 million users and roughly 2 million enterprise accounts, which makes the move feel less like a startup experiment and more like a signal to the market about where product software executives think the labor curve is heading.
That is why the “AI-pilled” framing matters. In the optimistic version, it means rebuilding the company around AI: workflows, incentives, hiring, and daily execution all get redesigned with software agents in mind. In the cynical version—the one this story keeps circling back to—it becomes a convenient story for headcount reduction. “Efficiency” sounds cleaner than “we think software can do your job,” even when the latter is closer to the truth.
The danger is not that AI is useless. It is that executives are treating it like a wholesale substitute before proving it can reliably handle the work. That distinction matters. A tool that speeds up drafting, triage, or internal search is not the same thing as a system that can own outcomes across support, operations, finance, or product delivery.
For developers, the implication is pretty clear:
- Build for specific workflows, not vague demos.
- Design for task decomposition, because agents are only as useful as the steps you can isolate.
- Treat monitoring and observability as first-class features, not afterthoughts.
- Keep human-in-the-loop review for the parts of the workflow where failure is expensive.
There is also a cultural split inside the industry. The OnlyCFO framing is more enthusiastic: it argues that being AI-first means putting AI into culture, hiring, incentives, and daily work, while admitting that many companies only pretend to be far along. That version is less about layoffs and more about organizational transformation. But the TechCrunch conversation suggests a darker reality: some firms may be using “AI agents” as a fresh label for an old corporate habit—cut first, justify later.
My read: the most overconfident companies are not the ones using AI aggressively. They are the ones confusing visibility with understanding. If a workflow looks automated on a slide deck, that does not mean it is actually automatable in production.
The real test for any AI strategy is not whether it sounds modern. It is whether the company can still deliver quality when the easy tasks are gone and the edge cases are all that remain.
