After rolling out roughly 900 plant cameras and a system that ingested engineering design specs to catch defects and supply disruptions, Ford brought veteran quality and safety staff back and acknowledged that AI 'is only as good as the information you use to train it.'
Ford is bringing back roughly 300 veteran quality and safety engineers after an internal AI quality system rolled out across its plants failed to match their judgment. The system paired about 900 AI-powered plant cameras with an engine that ingested engineering design requirements, intended to flag defects at the source and to help anticipate supply disruptions.
That AI bet now sits on the record. Charles Poon, Ford's vice president of vehicle hardware engineering, told reporters that "over prior years, we didn't pay as much attention as we should have to the experience of our most knowledgeable engineers." He added that "artificial intelligence is a fantastic tool, but it's only as good as the information you use to train it." Poon's framing is, in effect, an admission that the rehiring of roughly 300 veterans is the corrective to that mistake.
The cameras and the ingested-specification engine ran up against the kind of judgment experienced engineers make when written specifications run out. A camera plus a design document captures what is enumerable: measurements, tolerances, named failure modes. It does not capture the unwritten heuristics a veteran engineer reaches for when a part looks wrong but does not violate any tolerance. Those heuristics are built over thousands of hours working with the same parts and the same assembly lines. They live in the engineer, not in the dataset.
Ford's bet was that the gap could be closed by feeding the system more structured data, including detailed design requirements, defect logs, and the visual stream from the cameras themselves. Poon's quote concedes the bet did not pay off fast enough for Ford to forgo the human expertise it had displaced. The rehired veterans are, in his framing, the reconciliation.
The episode also complicates a debate Ford's chief executive has already taken public. Jim Farley has publicly framed AI as a force that "will leave a lot of white collar people behind," a stance that read, before the reversal, as a warning aimed at knowledge workers outside the factory. The rehiring does not contradict that stance. It places a limit on it. The same wave of enterprise AI that Farley expects to displace white-collar work apparently cannot, on current evidence inside Ford's plants, substitute for the most experienced hands in a quality function. Poon's concession is a tacit admission that the value of those veterans sits in something the engineering organization had not managed to make explicit enough to train on.
Two concrete questions follow. First, can Ford convert the institutional memory the rehired veterans carry into structured inputs a future version of the camera system can absorb, or does the reversal imply a permanent hybrid in which AI flags and humans rule on edge-case defects. Second, whether manufacturers running equivalent rollouts read Ford's reversal as a tacit-knowledge ceiling or as a calibration miss that more data and longer training can close.
Reporting on the underlying facts is sourced to Bloomberg's coverage as relayed by Albawaba, with specific workforce and infrastructure figures treated as Bloomberg-asserted pending direct verification.