Charles Poon, Ford’s vice president of vehicle hardware engineering, was remarkably candid about the misstep. He admitted that Ford had assumed feeding design requirements into AI systems would naturally produce a high-quality product. It didn’t. Many of the company’s most experienced engineers had already left before their knowledge could be captured, and the AI tools lacked the depth of training and contextual awareness that those people carried with them.
It’s a story worth paying attention to because the approach to AI failed.
The golden roof on a shack
At Jendamark, our innovations director Yanesh Naidoo has been making this point for years. He calls it putting a golden roof on a shack: layering expensive, sophisticated AI systems on top of a fragile operational foundation that was never designed to support them.
The analogy is simple but it cuts deep. If your data is fragmented across disconnected systems, if your processes aren’t standardised, if the people who understand your production line have walked out the door, then no amount of AI investment is going to save you. You’ll spend the money, you’ll get the dashboards, and the shack will still come down.
Ford’s experience is a textbook case. The company deployed AI across its industrial system with real ambition. But the foundation underneath, the process knowledge, the training data, the contextual understanding of how and why things go wrong on a line, wasn’t there to support it. The AI cameras could see, but they couldn’t understand the way a veteran inspector with 20 years of muscle memory could.
Yanesh has spoken about this extensively, from his newsletter on why smart factories still fail to his Thursday Thoughts video series, where he breaks down the fragmentation tax, which is the hidden 30% cost penalty that manufacturers pay when they try to stitch together disconnected point solutions instead of building from a unified foundation.
Go to the Gemba first
There’s another angle to the Ford story that deserves attention: the decision to bring the veteran engineers back.
Ford didn’t just rehire these people to do inspections. According to the company, they now act as internal auditors running mandatory weekly design reviews. They mentor junior staff. And critically, they train the AI systems, feeding them the contextual knowledge that was missing in the first place.
This mirrors a principle that came up on a recent episode of the Disrupted Factory podcast, where guest Vitor Ferreira spoke about the first step in any lean implementation: observe without interfering. Go to the Gemba, the place where the work actually happens, and watch. Don’t correct the operator. Don’t jump to solutions. Just capture reality.
Ferreira made the point that if you ask an engineer for a process flow, they’ll draw how it should work. Ask a manager, and they’ll draw how they think it works. Go to the floor, and you’ll see a third version: what’s actually happening.
Ford learned this the hard way. The AI was trained on design requirements (how the process should work.) But it wasn’t trained by the people who knew how it actually worked. That gap between the spec and the shopfloor is exactly where quality problems live, and it’s exactly where AI struggles without the right human input.
The foundation comes first
None of this means AI doesn’t work in factories. It does, but only when it’s built on the right foundation.
That’s the approach we’ve taken with ODIN Workstation, which we describe as an operating system for assembly lines. Rather than bolting AI onto disconnected systems, it starts with configuration, a digital representation of what the line should do, station by station, device by device. Then it enforces that process through worker guidance, hardware integration, and AI-powered computer vision. And then it validates: comparing what was planned against what actually happened.
The result is contextually rich, high-fidelity data. The kind that AI actually needs to be useful. Not fragmented signals from disconnected systems, but a complete picture of planned vs. actual across every station.
That’s the difference between a golden roof on a shack and a properly built structure. Ford is now learning what it takes to retrofit that foundation after the fact by rehiring the people, rebuilding the knowledge base, retraining the systems. It’s a far more expensive path than getting the foundation right from the start.
The lesson for manufacturing leaders
Ford’s story ended well as the company topped the JD Power 2026 Initial Quality Study for the first time since 2010. But it took three years of hiring back the people they’d lost, replacing senior leaders across engineering and supply chain, and fundamentally rethinking how AI fits into their operation.
The lesson isn’t that AI is overhyped. It’s that AI without foundation is just an expensive experiment. Before you invest in the golden roof, make sure the walls can hold it up.
