Should I trust the AI or my own judgement?

It is a real decision, the kind with your name on it, and the screen in front of you is confident. The model has run the numbers and made its recommendation. Twenty years of doing this work is telling you something else. And you sit there, cursor hovering, thinking: I no longer know which voice to believe, the model or the instinct that has carried me this far.

Here is the answer, before the diagnosis. You are not meant to pick one. The question is not whether to trust the AI or your gut instinct; it is which one deserves more weight in this specific decision, and that is a skill you can build rather than a coin you have to toss. Intuition and analysis were never rivals. They are two instruments reading the same situation, and the leaders who win the next decade are the ones who learn to calibrate between them on purpose.

Why does the model's voice suddenly feel louder than my own?

Because, for many leaders, it already is. SAP and Wakefield Research surveyed 300 C-level executives at US companies with at least a billion dollars in revenue. Forty-four percent said they would override a decision they had already planned to make once AI handed them a different read. Thirty-eight percent would let AI make the call on their behalf. And seventy-four percent said they trust AI for advice more than they trust their own family and friends. Sit with that last one. The machine's voice is not just in the room. For three out of four executives, it is winning.

That is the symptom. The problem underneath it is that the trust is unstable and openly conflicted. Salesforce found in 2025 that 76% of leaders feel a rising pressure to back their decisions with data, even as their own faith in that data falls. So you reach for the model to defend the call, then quietly doubt the thing you reached for. That is not a character flaw. That is a person being asked to outsource a judgement muscle they spent a career building, with nothing yet in place to tell them when the outsourcing is wise and when it is a mistake.

When should my gut outrank the model?

This is the part with a real answer, and it comes from the two people who spent careers arguing about it. Gary Klein studied expert intuition that saves lives. Daniel Kahneman studied the biases that wreck it. (Kahneman, who died in 2024, won the Nobel prize in economics for showing how predictably human judgement goes astray.) In 2009 they sat down to resolve the disagreement and found they largely agreed. Expert intuition can be trusted under two conditions: the environment has to be regular enough to offer valid cues, and you have to have had a long run of rapid, unambiguous feedback to learn those cues. A firefighter sensing a floor about to collapse, a trader who has lived through cycles, a retailer who can feel a store's week from the doorway. Where those conditions hold, your gut is data of the highest order.

Where they fail, confidence becomes the trap. A novel market, a one-off bet, a situation that has never given you clean feedback: there your certainty tells you nothing about whether you are right. So the calibration question is not "do I feel sure?" It is "have I earned the right to feel sure about a situation shaped like this one?" That single reframe changes which voice you weight, and it is teachable.

What does over-trusting the AI actually cost?

The same amount your gut costs when it overreaches, just in the other direction. In a 2024 computational pathology study (computational pathology is the AI-assisted reading of tissue images to diagnose disease), AI decision support raised overall diagnostic performance, which is the case for using it. It also produced a 7% automation-bias rate, where pathologists who had been correct on their own overturned their right answer to follow wrong AI advice. (Automation bias is the well-documented human habit of trusting a machine's suggestion over your own correct read.) Time pressure did not make the bias more frequent, but it sharpened the cost, with experts leaning harder on the negative AI suggestions and performing worse. Read that as a leadership warning, not a medical one. The model does not have to be wrong to cost you. It only has to be confident at the moment you are tired and rushed.

There is a humbling footnote here about the human side of that table. In a study of Israeli parole boards, favourable rulings drifted from roughly 65% at the start of a session towards almost zero by the end, then returned to 65% after a food break. The exact magnitude has been contested by later researchers, so hold it as suggestive rather than settled. But the direction is sobering: the same expert judge, same case quality, swung by something as crude as the hours since lunch. The instinct you are weighing against the model is running on hardware that gets tired. The model never breaks for lunch. Neither of those facts settles the decision; together they tell you the contest is closer than it feels.

How do I build the calibration, not just feel the tension?

Start by treating the two voices as colleagues who disagree well, then steward the disagreement on purpose.

1. Name the domain first. Before you read the model's output, ask whether this is home turf or a genuine first. That single classification decides which voice carries more weight, and it stops the screen from setting the frame for you.
2. Let each voice speak before they meet. Form your own read, then open the model's. Reversing the order lets the AI anchor you, which is exactly how that 7% of correct judgements got overturned.
3. Interrogate agreement as hard as conflict. When the model confirms what you already wanted, that is the moment to slow down. Relief is not evidence.
4. Check the state of the instrument. If you are rushed or running on empty, defer the irreversible call. The parole-board drift is a warning about your own afternoons.

That last point is where the deeper work lives, and it is more physical than most leaders expect. Higher resting heart rate variability, the vagally-mediated kind (vagally-mediated means driven by the vagus nerve, the body's main brake on the stress response), is associated with stronger executive function and cognitive control; the neurovisceral integration model (the theory that the heart and the thinking brain regulate each other) reads it as a marker of how well your prefrontal cortex is regulating in the moment. Plainly: a calm, coherent nervous system makes better judgements than a frayed one. The voice you bring to the model is only as clear as the state you are in when you bring it. Most leaders are trying to install new software on broken hardware. The bottleneck is no longer the technology; it is the quality of the human judgement meeting it.

This is also why the technology alone keeps underdelivering. Deloitte's 2026 Global Human Capital Trends study, drawing on more than 9,000 business and HR leaders across 89 countries, found that 60% of executives now regularly use AI to support their decisions, yet only 5% say they manage that well. Read the gap between those two numbers slowly. Adoption has raced ahead; the discipline of weighting the model against your own judgement has barely started. The same study found that 59% of organisations still take a purely tech-focused approach, and that those organisations are 1.6 times more likely to miss the returns they expected from AI. The companies pulling ahead are not the ones with the better model. They are led by people who have learned to hold the model and their own knowing in the same hand, and to weight them on purpose. That is the capability worth building this year.

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.