What is 'AI psychosis' and why is Aaron Levie warning about it?

AI psychosis is Box founder Aaron Levie's term for executives confidently declaring AI can replace jobs they barely understand. The decision-makers loudest about automation are often the furthest from the actual work, creating a dangerous blind spot. Levie argues these leaders overestimate AI's real coverage because they never mapped the implicit judgment, exception handling, and tacit knowledge inside the roles being cut. The result: rushed mass layoffs that look efficient on a spreadsheet but degrade organizational capability. His critique targets decision quality, not AI adoption itself.

How bad are 2026 AI-driven tech layoffs compared to 2025?

Before mid-2026, tech layoffs have nearly matched the entire 2025 total. ClickUp's recent announcement cutting 22% of its workforce — explicitly replacing those roles with AI Agents — is one of the most visible examples. The trend is accelerating because executives now treat 'AI Agent' as a justification line in restructuring plans. The pace matters: when cuts outrun real automation capability, companies lose institutional knowledge faster than they recover productivity, and rehiring at higher cost becomes common within 6–12 months.

How do I avoid overestimating what AI Agents can actually replace?

Interview the people doing the job before designing any replacement plan. Ask one specific question: 'What do you know that nobody else knows?' That surfaces the implicit judgment, edge cases, and cross-team coordination that never appears in a job description. Map the visible workflow steps, then map the invisible decisions on top. AI Agents handle the visible layer well but fail on tacit knowledge and exception handling. If you cannot describe both layers clearly, you are not ready to automate — you are ready to fail expensively.

What's the most common mistake companies make when deploying AI Agents?

Skipping deep requirements analysis with actual executors. Teams pick the tech stack, design the agent flow, and ship — without ever sitting with the person whose work is being automated. This guarantees the agent covers the documented 60% of the role and misses the undocumented 40% that holds everything together: customer context, judgment calls, escalation instincts, and informal coordination. The fix is not more AI investment. It is treating automation as a requirements problem first, an engineering problem second.

Is Levie saying companies should stop adopting AI?

No. Levie explicitly supports AI adoption — his warning targets reckless replacement decisions made by leaders disconnected from the work. The distinction matters. Augmenting workflows, automating well-understood tasks, and building agents around documented processes are all reasonable. Cutting 22% of headcount based on a vendor demo and an executive hunch is not. Treat AI as a capability multiplier where the underlying work is already deeply understood. Where understanding is shallow, slow down, study the role, then automate the parts you can defend.

Who should read Levie's warning most carefully?

Three groups: executives planning AI-driven restructuring, product leaders designing internal agent workflows, and founders building AI replacement products for enterprise buyers. For executives, the lesson is decision hygiene — verify your understanding of a role before eliminating it. For product leaders, design agents around real workflow interviews, not org charts. For founders selling AI replacement tools, your customers' overconfidence is your churn risk: when their layoffs fail, they blame your product. Build in onboarding that forces requirements discovery before deployment.

What happens when companies become too AI-pilled?

This article is a deep-dive from JudyAI Lab — an AI engineering playbook series with 100+ published guides, 5,000+ weekly readers across 60+ countries, focused on the practical side of running AI agents, trading systems, and content pipelines in production.

📰 TL;DR

Box founder Aaron Levie recently called out what he calls “AI psychosis” — the phenomenon where executives who greenlight “AI can replace this job” are often the ones who know the least about what that job actually entails. He’s warning that this decision-making blind spot is spreading across tech — decision-makers, overconfident in AI’s potential, are rushing to implement mass layoffs without truly understanding workflows, role nuances, or the human judgment required.

In a concrete case, collaboration platform ClickUp just announced cutting 22% of its workforce, explicitly stating it will use AI Agents to take over those functions. This wave of layoffs has pushed 2026’s total tech layoffs to nearly match all of 2025 — before even reaching the mid-year point — showing AI-driven workforce reduction is accelerating.

Levie’s core argument isn’t against AI adoption, but rather a warning: companies rushing to replace human labor with AI lack deep understanding of the work itself. When decision-makers aren’t familiar with the actual complexity of the roles being replaced, they tend to overestimate AI’s real coverage capability, ultimately hurting organizational efficiency. This “over-AI’d” thinking is becoming a new management risk in Silicon Valley. Full interview available at the source link.

💬 JudyAI Lab Take

The “AI psychosis” Levie pointed out is a red flag worth every AI implementer paying attention to: the execs loudest about “AI can replace this job” are often the ones most unfamiliar with that role — that’s the real decision blind spot.

ClickUp’s 22% headcount reduction replacing roles with AI Agents has already pushed 2026 tech layoffs near matching all of 2025 — before midyear. There’s a wake-up call for the AI builder community here: the环节 where automation design fails most isn’t tech selection, but insufficient understanding of the “work being automated.” When designing Agent flows or workflows, skipping deep interviews with actual executors easily leads to overestimating AI’s coverage — automating visible steps while missing a lot of implicit human judgment and exception handling. Levie’s critique is essentially a requirements analysis problem: not understanding the true complexity of a job means more AI investment can lead to bigger organizational efficiency losses.

Before planning any AI replacement plan, talk to the people actually doing that job first. Ask them “what do you know that nobody else knows?” — that’s often exactly where AI falls shortest.

📰 TL;DR#

💬 JudyAI Lab Take#

References#

Get our weekly AI digest:

📰 TL;DR

💬 JudyAI Lab Take

References