# Fast Company | BYLINE | Can AI learn a gut feeling? - July 1st, 2026 ATLANTA – ( [_Fast Company_](https://www.fastcompany.com/91567064/can-ai-learn-a-gut-feeling)) A structural engineer I once worked with could glance at a set of drawings for a few seconds and say, “Something’s off.” Not wrong in a way she could name on the spot, just off. She’d sit with the unease, and 20 minutes later she’d surface it: a load path that didn’t close, or an assumption that didn’t hold. The knowing came first. The explanation came later. Every expertise-heavy field has a version of her. The radiologist who senses a mass before pointing to it. The fire chief who, in a case documented by [researcher Gary Klein](https://mitpress.mit.edu/9780262611466/sources-of-power/), ordered his crew out of a building seconds before the floor collapsed into a basement fire he never saw. He couldn’t say what was wrong. He only knew that something was. It looks like magic, but it isn’t. It may not be teachable either, and that distinction is about to get very expensive. ### **The Experience Exodus** And the people who hold that experience are leaving. In 2025, about [11,400 Americans turned age 65 every day](https://www.prnewswire.com/news-releases/the-us-has-reached-the-peak-of-peak-65-its-time-to-apply-retirement-readiness-lessons-from-the-boomer-experience-302360086.html), the largest such wave in history. In my own profession the shift is already visible: [A third of all U.S. architects are baby boomers](https://www.ncarb.org/sites/default/files/NBTN-2024.pdf). In 2024, the [number of licensed architects dropped 4%](https://www.ncarb.org/nbtn2025/state-of-licensure), while 13% are already past retirement age. And a [2025 APQC survey](https://www.apqc.org/about-apqc/news-press-release/apqc-study-warns-looming-great-retirement-crisis-highlights-role-ai) found that 92% of organizations don’t consistently capture the know-how of experts nearing retirement. We are losing the people who hold the hardest-to-replace judgment faster than we are replacing them. The obvious hope is that [artificial intelligence](https://www.fastcompany.com/section/artificial-intelligence) can absorb that judgment before it leaves. Whether it can depends on what goes into judgment and how it’s defined. ### **The Thing We Can’t Put Into Words** The philosopher Michael Polanyi described the obstacle 60 years ago in seven words: “We can know more than we can tell.” He called it tacit knowledge, the competence that resists being written down. In my own work, I could hand a new hire every building code and past project file and still not transfer the instinct of my long-time engineer. That’s the core difficulty. You can’t capture what the expert themselves cannot articulate. So, can a machine learn it anyway? The obvious answers pull in opposite directions. ### **The Optimist’s Case** Maybe intuition is less mysterious than it feels. The economist Herbert Simon reduced it to a sentence: Intuition is “nothing more and nothing less than recognition.” The expert has simply seen so many patterns that the right one fires instantly. And pattern recognition at scale is precisely what modern AI does best. It’s hard to argue with the evidence. [DeepMind’s AlphaFold](https://deepmind.google/science/alphafold/) cracked protein structures that had defied expert intuition for 50 years. [Mayo Clinic’s AI-enabled electrocardiogram](https://www.nature.com/articles/s41591-018-0240-2) read an ordinary ECG and flagged heart dysfunction cardiologists couldn’t see. If intuition is recognition, the machine is already learning to recognize. ### **The Skeptic’s Case** An expert’s judgment, however, carries something the AI pattern can’t: accountability. My engineer signs her name to the drawing; if the building fails, that’s on her. That stake shapes the instinct in a way no model shares. Confidence without consequence is dangerous. Boeing’s 737 MAX relied on an automated system called MCAS that was fed by a single sensor and never disclosed to pilots. When the sensor failed, the system forced the planes into nosedives the crew couldn’t override. [Two crashes, five months apart, killed everyone on board.](https://democrats-transportation.house.gov/news/press-releases/after-18-month-investigation-chairs-defazio-and-larsen-release-final-committee-report-on-boeing-737-max) Less dramatic, but equally serious, is what AI does to human experts. For example, when a tool offered wrong readings, [radiologists with more than 15 years of experience saw their accuracy collapse from 82% to below 46%](https://www.rsna.org/news/2023/may/ai-bias-may-impair-accuracy). The sense that “something’s off” is built from having been burned, from failure, from liability, and from years on the jobsite and in the work. You can’t separate the instinct from the experience that built it. ### **The Reframe** Replicating intuition is probably the wrong thing to chase. The useful question is whether AI can transfer enough of intuition’s output to extend it. There’s real evidence it can. [In a study of more than 5,000 customer-support agents](https://www.nber.org/system/files/working_papers/w31161/w31161.pdf), access to an AI assistant raised the productivity of novices by 34% while barely moving the veterans. The researchers concluded that the system effectively passed along the tacit knowledge of top performers to the people still climbing the curve. The AI didn’t become the expert. It carried a compressed trace of the expert to someone who needed it. In architecture, that might look like a junior designer running a set of drawings past an AI before the principal ever sees them, catching the kinds of issues that used to surface only in review. We probably can’t bottle the thing my engineer couldn’t explain. We might be able to capture enough of her decisions, the catches and the second looks, to give the next generation a head start instead of a blank page. For leaders, that points to a practical shift in what to capture. Most knowledge programs record outputs: the finished drawing, the closed file, and the final memo. But the transferable value lives earlier, in the process. It’s the markup where a principal looked at a junior designer’s elevation and moved the building entry because afternoon sun would blind anyone walking out. The final site plan just shows a door in the right place. Most firms archive the clean version. The clean version is the least useful one. Build your systems, and the AI tools you adopt, to capture the reasoning and the redlines, not just the result. Don’t wait for the retirement party. The time to capture how your best people decide is while they’re still deciding. The question isn’t whether AI can learn a gut feeling. It’s whether your firm is capturing the reasoning behind your best people’s decisions before those people leave. * * * ABOUT THE AUTHOR Sandeep Ahuja is the co-founder and CEO of cove, the leading Vertical AI Services company for building design, dedicated to transforming architecture for a sustainable future. With over a decade of experience, she has revolutionized the design and construction process by promoting seamless collaboration, shared knowledge, and accountability.