"We had this idea that if we just adopted AI, high-quality products would follow." That line, delivered by a Ford executive in an interview with TechCrunch, is understated but heavy with implication. After rolling out AI, Ford fell short of its own quality targets, and to close the gap, the company started calling its retired veteran engineers back in. The so-called "gray beards" — engineers who'd spent decades on the floor — traded quiet retirement for coveralls once again.
On the surface, this reads like a story about AI failing. Look closer, though, and a different question emerges. Ford is one of the oldest manufacturers in the world, and it was hardly reluctant about adopting AI. The reason it had to call field experience back in has less to do with how mature the technology is than with the nature of experience itself as an asset.
The Gap AI Couldn't Fill
Ford's reasons for embracing AI aggressively were clear enough: cut labor costs and speed up product-design review and quality control. AI was put to work analyzing drawings, catching design errors, and processing production data. At first, the approach looked like it was working.
The trouble showed up on the factory floor. AI-reviewed or AI-generated output kept failing to meet real manufacturing standards. Designs that passed muster on paper caused problems once they hit the assembly line. How a given material behaves at a certain temperature, the unwritten standards built up over years of working with the same suppliers, the practices everyone on the floor simply follows even though they appear nowhere on a drawing — none of that knowledge had ever made it into AI's training data. It couldn't have. It was never documented in the first place.
That's why the retirees came back. What they carried with them was judgment that exists nowhere in a company manual: why you'd choose one of two similar-looking designs over the other, which tolerances are fine and which ones fail on the floor — intuition built from decades of things going wrong. It's the kind of knowledge that resists being written down.
Ford's story isn't confined to manufacturing. Consider what happens at an advertising agency when a senior director who understood a client's internal decision-making structure leaves, and an AI-driven campaign-analytics tool takes her place. Consider what kind of judgment disappears from a publishing house when the editor who could read a manuscript and intuit the market's response moves on, leaving only an AI reader-analysis tool behind. Consider what quietly goes missing at a tax practice when the veteran who knew, in his bones, how to approach each type of taxpayer is gone, and only automated filing software remains. The knowledge these people carried is the same kind. AI struggles with it not because of some performance shortfall, but because that knowledge was never data to begin with.
The Claim That More Compute Will Fix It
It's tempting, and logical enough, to read this episode as a story about AI's technological immaturity. Today's models, the argument goes, still struggle with deep, unstructured, domain-specific knowledge, and as model performance keeps improving, the problem Ford is running into now will resolve itself over time. It's a fair point: over the past three years, AI model performance has outpaced many predictions, with especially striking gains in coding, math, and logical reasoning. Seen this way, Ford's move looks like a temporary retreat — once the technology matures enough, the thinking goes, the company will be able to hit the same quality bar without veteran engineers.
But that argument skips an important premise: before a model can learn something, that knowledge has to exist as data in the first place. Nothing that Ford's veteran engineers know has ever been written down anywhere. It was never recorded, so it was never learned — and no matter how much model performance improves, you can't pull knowledge out of data that doesn't exist. A stronger model just means better handling of data that already exists. It doesn't mean compensating for experience that has never been turned into data at all.
That produces a second paradox. The more ground AI can competently cover in structured work, the larger the remaining, harder-to-automate share becomes by comparison. As AI takes on more of what can be handled, the value of experience and judgment in what's left over only rises. As the technology advances, unstructured experience doesn't become obsolete — it becomes scarce.
HR research on organizational knowledge transfer and capability management has been pointing this out for years: some of what a skilled practitioner knows can be passed on through training and manuals, but some of it can only be acquired through direct, on-the-job experience. Push AI adoption without distinguishing between the two, and the cost of making that distinction later — after the fact — comes back larger. Major South Korean corporations went through a similar pattern in 2023 and 2024. More than a few of them thinned out their ranks of experienced middle managers while rolling out AI tools, only to discover, belatedly, that no one was left in the organization who could set the standard for judging whether AI's output was any good.
Where Is Your Gray Beard?
Reframe Ford's story from the vantage point of a Korean solo entrepreneur, freelancer, or one-person PM operation, and the question changes. It's no longer "how well can AI do what I do?" It's "is the judgment I'm handing over to AI actually my core asset?"
It's efficient for a designer to let AI generate the first draft of a concept. But if she starts relying on AI's suggestions to decide which direction fits a client's organizational culture, or which tone actually lands with a brand's buyers, the experience that once sharpened that instinct stops accumulating in her. It's reasonable for a marketer to use AI to spin up content ideas fast. But hand over the judgment of which message actually moves a given audience to AI analysis as well, and that instinct atrophies. Look closely at why Ford's quality targets collapsed, and the real gap wasn't in the AI tools' performance — it was the absence of people who could tell whether AI's output was right or wrong.
Worth checking: think about which of the tasks you're currently handing to AI are the ones that have actually built your judgment over the past several years. Setting expectations with clients, negotiating price, prioritizing projects, choosing partners. Getting data or a first draft from AI to work from is fine. The ability to judge whether that draft holds up is something you need to hold onto yourself.
If it's been six months since you started using AI tools in earnest, it's worth pausing to check whether your own judgment has gotten sharper or duller over that time. If leaning on AI as a reference has made your eye more refined, that's the right direction. If you find decisions getting harder without AI, or your own instincts shrinking, you're quietly losing something else underneath the efficiency gains.
Ford only found out after the fact. It had to talk retirees into coming back, and it took time to fix the quality problems. That cost more than it would have to run AI and experience together from the start.
It's worth asking what your own gray beard is — and whether it's still alive in you.
