Tech CEOs May Be Overestimating AI’s Immediate Impact

The tech sector is experiencing a wave of layoffs that rivals the entire previous year, with 115,430 workers let go from 152 companies in the first five months of 2026, according to Layoffs.fyi. A sizable share of those cuts are being justified by AI‑driven efficiency narratives. One voice cutting through the noise is Box founder Aaron Levie, who recently posted on X that CEOs are prone to what he calls AI psychosis—a tendency to over‑promise on what artificial intelligence can deliver because they are removed from the day‑to‑day work that actually creates value.

The core of Levie’s argument

Levie points out three blind spots common among top executives:

1. Distance from the implementation layer – CEOs may prototype an AI tool, sign a contract, and assume the technology will replace large swaths of human effort without seeing the debugging, model‑tuning, and data‑curation that keep it functional.
2. Misreading the limits of automation – Many tasks still require nuanced judgment, especially when contracts contain “sneaky” clauses or when code must be vetted for hallucinated library calls.
3. Underestimating the new bottleneck – Even if AI generates output at scale, executives become the gatekeepers who must approve, correct, and prioritize that output.

Levie is not an AI skeptic; he regularly shares optimistic posts about “headless software” and backs AI‑focused startups as an angel investor. His recommendation is pragmatic: CEOs should use AI heavily themselves, experience its failures first‑hand, and develop a balanced view of both upside and the work that remains.

Real‑world examples of the hype‑to‑reality gap

• ClickUp’s CEO Zeb Evans announced a 22 % staff reduction after deploying roughly 3,000 internal AI agents. He frames the move as a shift toward a “100x org” where humans supervise agents rather than perform routine tasks. The claim that such a model will automatically multiply output has little empirical support.
• A meta‑analysis published in California Management Review (Oct 2023) found no robust correlation between AI adoption and aggregate productivity gains across firms.
• Research from the National Bureau of Economic Research (Mar 2024) did detect a productivity lift, but described it as a productivity paradox: perceived gains outpace measured ones.
• MIT researchers evaluating thousands of autonomous agents concluded that, today, many agents still fall short of human‑quality work. Their forecast suggests a 80 %–95 % success rate on text‑heavy tasks by 2029, which they deem “minimally sufficient.”
• A Harvard Business Review article highlighted a new bottleneck: as AI democratizes content creation, executives become the choke point for reviewing and authorizing that output.

Why the disconnect matters

When CEOs treat AI as a plug‑and‑play solution, they risk two outcomes:

1. Organizational chaos – Over‑automation can create a flood of low‑quality deliverables that require extensive human correction, stretching managerial capacity.
2. Talent attrition – Premature layoffs erode institutional knowledge and can damage a company’s reputation, making it harder to attract top engineers who understand both AI and the underlying domain.

The data suggests that AI is still a tool, not a replacement for skilled labor. Companies that integrate AI thoughtfully—by pairing agents with subject‑matter experts, iterating on model performance, and measuring real‑world impact—are more likely to see sustainable gains.

What CEOs can do now

• Hands‑on experimentation: Spend time with the same AI systems their teams use, noting false positives, hallucinations, and edge cases.
• Define clear metrics: Track not just cost savings but also error rates, time‑to‑resolution, and employee satisfaction after AI rollout.
• Invest in upskilling: Enable staff to become “AI supervisors” rather than merely “AI users,” ensuring the human layer adds value.
• Communicate transparently: When announcing AI‑driven changes, be realistic about timelines and the need for ongoing human oversight.

By grounding expectations in measured outcomes, CEOs can avoid the pitfalls of what Levie describes as AI psychosis and steer their organizations toward genuine, incremental improvement.

---

Sources: Layoffs.fyi, California Management Review, NBER Working Paper, MIT research on autonomous agents, Harvard Business Review article on AI bottlenecks.