Ford spent three years leaning on an AI-driven quality system to catch defects before they reached buyers. It didn’t work. By 2023, Ford ranked near the bottom of J.D. Power’s quality survey — its worst showing in 16 years. The fix wasn’t more AI. It was rehiring 350 veteran engineers — employees call them “gray beards” — who’d spent decades on the factory floor before earlier layoffs pushed many of them out. This year, Ford topped J.D. Power’s rankings for the first time in 16 years.
Here’s the mechanism, and it’s a tacit knowledge problem, not an AI failure. AI systems learn from what’s been written down, measured, structured. Decades of shop-floor judgment — the kind that lives in someone’s hands, not a spreadsheet — never got captured before the people who held it left. You can’t feed a model data that was never collected. Ford’s AI wasn’t undertrained; it was starved of the one input that mattered.
That’s the exact question that decides which slot a stock belongs in: can AI actually generate this company’s core value, or does it require something AI cannot manufacture from data alone? A live sports league, a name you’ll pay more for on reputation alone, a piece of institutional judgment nobody wrote down — all of them survive precisely because they resist formalization. Ford’s failure is the negative-space proof of the same rule: the moment you can fully formalize a knowledge base, you’ve also made it replaceable — and if the people holding it leave first, you get neither version. The AI moat isn’t disappearing, it’s relocating — tacit knowledge is one of the places it’s relocating to.
My take: “Some knowledge only exists while someone’s still holding it.”
Not advice. Just how I see it.