Has AI quietly distorted how you see your own business?

I caught myself promising the board things about AI that, in the cold light of day, I am not sure are even true. You know the moment. You are mid-sentence, and you hear yourself say it: AI will take a third off our cost base by Q3, and we will redeploy the rest into growth. The room nods. You feel sharp, ahead of it, finally the leader who saw it early. Then, that night, over the spreadsheet, a quieter voice arrives, and it asks whether you can actually stand behind a word of it.

If you have had that moment, your judgement is intact. You are experiencing something with a name now, and a mechanism behind it. The short version: the model talks like it has already done the work, and that fluency is contagious. It is easy to mistake a confident demo for a delivered result, and to carry that confidence upward into a claim the business cannot yet honour. The fix is more rigour, not less ambition: a practice of reality-testing, run deliberately, before the claim leaves your mouth.

What is "AI psychosis" in leaders, and is it real?

It is real enough that the people building these tools are warning about it. Aaron Levie (chief executive of Box, the enterprise file-sharing company) popularised the phrase "AI psychosis" for executives in late May 2026, and his diagnosis is precise: chief executives are uniquely prone to it because they sit furthest from the last mile of work that actually generates the value. They play with the model, they see the happy path, and they do not picture the next ten or twenty things that have to happen for the result to hold. His worked example reads almost word for word like the board-meeting moment above. The leader says, look, I generated a contract. And the truth is, you did not verify every term before it went to the other side (Fortune, 2026).

That is the shape of it. Not delusion. A coherence failure: the head makes a claim that the heart, and the numbers, have not yet matched. And the data says it is widespread. In a global study of more than 48,000 people across 47 countries, 66% said they rely on AI output without checking its accuracy, and 56% admitted to making mistakes in their work because of it (KPMG and the University of Melbourne, 2025). The confident-fluency-detached-from-reality dynamic is documented, not anecdotal. You are not the exception. You are the pattern.

Why does AI's confidence distort how I see my own business?

Because the seduction operates at the level you trust most: your own pattern-recognition. You built a career on reading situations fast and being right. AI hands you outputs that feel like that same fast, expert read, except the model has no skin in your P&L and no idea what the last mile costs. So the claim inflates, and it inflates upward, into the very rooms where your credibility is the currency.

The numbers underneath the hype are sobering, and they are the part the demo never shows you. In Deloitte's latest enterprise survey, 74% of organisations say they hope to grow revenue through their AI initiatives in future, while only 20% report that they already are; just 34% say they are truly reimagining the business rather than tidying up existing processes (Deloitte, State of AI in the Enterprise, 2026). The technology works in the demo. The business case lives somewhere the demo cannot reach.

Look at the second row closely, because it is the one that ends up in your board narrative. McKinsey found 88% of organisations now use AI in at least one function, up from 78% a year earlier, yet only 39% report any EBIT impact at the enterprise level, and among those, most say it is less than 5% (McKinsey, 2025). The adoption is real. The arrival is mostly still a promise. And the gap between the two is exactly the space where a confident claim becomes an exposed one.

How do I reality-test an AI claim before it reaches the board?

You build a practice, and you run it every time, especially when the claim feels obviously true. That is the tell. The most dangerous claims are the ones that arrived frictionlessly. Reality-testing is the type of intervention that addresses this whole class of problem: it reintroduces the friction the model removed, on purpose, in the places that protect your credibility.

1. Separate the demo from the deployment. Write down what you actually saw the model do, in one sentence. Then write the ten or twenty things that have to happen for that to hold at scale. The distance between those two is your real claim.
2. Name the last-mile owner. For every AI promise, name the human who closes the gap between output and outcome. If you can name them, the value is on its way; if you cannot, it is still a forecast wearing the clothes of a result.
3. Check the claim against the profit and loss account. Ask one question before the board does: where, exactly, does this surface in EBIT, and by when? If the honest answer is "soon, somewhere," it is ready for the workshop, not the boardroom.
4. Decide from a settled state, not a stimulated one. The confident claim usually lands late in a long day, when your capacity for deliberate thinking is depleted and your brain reaches for the fast, intuitive shortcut. Make the big calls when you are clear, while you still have capacity in the tank.

That last step holds its weight. Decision-fatigue research shows self-control behaves like a depletable resource, and as it wanes the brain shifts from analytical thinking toward intuitive shortcuts; the effect is real enough that, as a long session wears on, judges grant parole less and physicians prescribe more unnecessary antibiotics (Pignatiello, Martin and Hickman, 2020). The state you decide from shapes the decision. Higher cardiac vagal tone, measured through heart-rate variability (the small beat-to-beat changes in your pulse, a marker of how well the calming branch of the nervous system is working), tracks with stronger prefrontal regulation and better executive performance under the neurovisceral integration model (the theory that heart and brain operate as one connected control system) (Thayer and colleagues, 2009); I treat that line of evidence as exploratory rather than settled, and worth your attention as a capability to build. Coherence, head and heart in alignment, is what keeps the confident claim and the true claim as the same sentence.

Here is the part that should settle you, because it points at where the real opportunity sits. The model is the easy part. Deloitte's enterprise survey names insufficient worker skills as the single biggest barrier to AI integration, and the leading response, in 53% of organisations, is educating the workforce rather than re-architecting the technology (Deloitte, 2026). The bottleneck is no longer the technology. So the credibility you are protecting in the boardroom is built in the same place the value is: in how your people and your processes change, which is the part you can see, lead and verify with your own eyes. The leaders who win the next decade are the ones who upgrade themselves first, starting with how clearly they see their own business.

About the author

Thomas Green

British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.