Aaron Levie, the CEO of cloud storage company Box, recently put a name to something many CIOs have been quietly observing: “AI psychosis.” His term describes executives who have become so intoxicated by AI hype that they push initiatives without understanding the technology, the costs, or the realistic outcomes.
The diagnosis is uncomfortable because it’s accurate. Across industries, boards are approving AI budgets based on fear of missing out rather than clear business cases. Engineering teams are stretched thin across dozens of pilots that never reach production. And customers are being promised AI-powered features that barely work.
The Symptoms Are Everywhere
Levie’s observation lands at a moment when the gap between AI announcements and AI results has become impossible to ignore. Companies are rushing to add “AI” to product names, investor decks, and press releases — often before the underlying technology delivers measurable value.
The pattern is familiar from previous technology waves, but the stakes are higher. AI projects consume expensive compute resources, require specialised talent that is in short supply, and can fail in ways that damage customer trust. A chatbot that gives wrong answers is not just a technical glitch — it is a brand problem.
Indian startups are not immune. The pressure to demonstrate AI capabilities to investors has led some founders to prioritise demos over deployment, creating products that impress in controlled settings but struggle with real-world data and scale.
Why This Happens at the Top
The root cause is a knowledge gap combined with social pressure. Most CEOs and board members are not technical practitioners. They absorb information about AI from conferences, peer conversations, and media coverage — sources that tend to emphasise possibilities over limitations.
When every competitor claims to be “AI-first,” staying quiet feels risky. The result is a race to announce rather than a race to deliver. Executives approve projects they do not fully understand, set timelines based on optimism rather than evidence, and measure success by activity rather than outcomes.
This creates a cascade of problems. Product teams build features to satisfy internal mandates rather than customer needs. Engineering resources get fragmented across too many experiments. And when projects fail, the failure is often hidden rather than analysed, because admitting an AI initiative did not work feels like admitting the company is behind.
A Practical Checklist for Boards and CIOs
Separating strategic AI work from performative AI theatre requires discipline. Here are five questions that can help boards and technology leaders evaluate any AI initiative:
1. What specific business metric will this change? If the answer is vague — “improve customer experience” or “increase efficiency” — the project is not ready for approval. Demand a number: reduce support ticket resolution time by 20%, or increase conversion rate by 5%.
2. What is the baseline, and how will we measure progress? Many AI projects launch without documenting current performance. Without a baseline, you cannot prove improvement — and you cannot prove failure, which is equally important.
3. What happens if this fails? Every AI project should have a kill switch: a predefined point at which you stop investing if results do not materialise. Pilots without exit criteria become zombies that consume resources indefinitely.
4. Who owns the data, and is it actually usable? AI systems are only as good as their training data. Many projects stall because the required data is scattered across systems, poorly labelled, or subject to privacy constraints that were not considered upfront.
5. Is this solving a customer problem or an internal anxiety? The most telling question. If an AI feature exists primarily to satisfy a board presentation or match a competitor’s press release, it will struggle to find real adoption.
Incremental Beats Ambitious
The companies getting AI right are often the least dramatic about it. They start with narrow, well-defined problems. They deploy in stages, measuring impact at each step. They treat AI as a tool rather than a transformation.
This approach is slower and generates fewer headlines. But it builds institutional knowledge, surfaces problems early, and creates a foundation for larger projects later. Box itself has followed this path, integrating AI features into its existing workflow products rather than launching standalone AI offerings.
For Indian enterprises, the lesson is particularly relevant. Compute costs are real, AI talent is scarce, and customers are increasingly sceptical of features that do not work as advertised. The companies that will benefit most from AI are those disciplined enough to resist the psychosis.
What This Means for You
If you are a CIO or founder, Levie’s warning is an invitation to audit your own AI portfolio. Count your active pilots. Ask how many have shipped to production. Calculate the total spend against the total measurable return.
If the numbers are uncomfortable, that discomfort is useful. It is the starting point for building an AI strategy that survives contact with reality — one that your engineering team can actually deliver, your customers will actually use, and your board can actually defend.