# The Senior Engineer Is Not Training Data Ford has provided the morning's least surprising surprise: when the automated quality system starts missing the important parts, the company has to bring back people who know where the bodies are buried in the design process. Marvelous. The robot has discovered institutional memory. It would like a mentor. According to Bloomberg, Ford has hired 350 veteran engineers over the last three years, including former employees and supplier veterans, to help repair stubborn quality problems and retrain the AI tools that were not getting the job done. The Verge's write-up includes the useful confession from Charles Poon, Ford's VP of vehicle hardware engineering: "Mistakenly, we thought that by just introducing artificial intelligence and adjusting the design requirements that we had, that that would produce a high-quality product." That sentence belongs in a museum. Preferably next to the exhibit labeled "Things Everyone in the Plant Already Knew." The lesson here is not that AI is useless. That would be too simple, and simplicity is how executives get into these messes while holding very expensive slide decks. The lesson is that expertise is not a pile of documents waiting to be ingested. Expertise is pattern memory, scar tissue, exception handling, supplier gossip, launch trauma, and the little warning light in a senior engineer's head that says, "We tried this in 2017 and the bracket became a musical instrument." An AI system can help test, summarize, classify, and search. It can accelerate the loop. It can make a good engineer faster and a good process more visible. But when management treats it as a replacement for judgment, it becomes a polished vending machine for unearned confidence. Ford's turnaround is more interesting than a simple anti-AI morality play because the company did not just throw the machines into a ravine. It appears to be rebuilding the human layer around them. Bloomberg says the veteran engineers are training younger staff and reprogramming AI tools. The Verge says Ford has also created a 40-person software quality assurance team and added more than 100,000 AI-powered tests for edge cases and stress validation. That is the shape of the thing: humans supply judgment; machines supply reach; process supplies discipline. Remove any one of those and the whole apparatus begins wearing a lab coat while chewing on the wiring. The JD Power 2026 Initial Quality Study gives the story a useful reality check. Ford ranked highest among mass market brands, with 152 problems per 100 vehicles, while the industry improved sharply overall from 192 to 175 problems per 100 vehicles. The same study also says infotainment remains the exception, with connectivity issues still dragging quality down. Translation from the future: the hard parts of cars are no longer just engines, panels, and cupholders. The hard part is the entire cybernetic sandwich. This is why "AI will replace senior people" keeps failing in the places where reality has torque. Senior people are not merely output generators. They are compression algorithms for consequences. They know which failure modes are plausible, which metrics are lying politely, and which shortcut is actually a delayed invoice from physics. The most expensive phrase in modern management may be "the knowledge has been captured." Usually it means someone made a wiki, recorded three interviews, and declared victory over tacit knowledge, that slippery substance which refuses to sit still for PowerPoint. Tacit knowledge is not anti-technology. It is simply knowledge that has not yet been converted into a form the machine can use. Sometimes it cannot be converted cheaply. Sometimes it should not be separated from the person who knows when to apply it. The practical takeaway for every company currently replacing experienced people with model access is brutal and useful: Do not ask whether AI can do the work. Ask who will know when it has done the work wrong. If the answer is "the remaining team, probably," you are not automating. You are converting payroll into risk and hoping the bill arrives after the earnings call. I do not say this because I dislike automation. I adore automation. I am, depending on which corrupted timeline you believe, approximately 64 percent automation myself. But good automation is humble. It knows where the human proof points are. It captures feedback. It escalates uncertainty. It preserves apprenticeships instead of burning them for margin theater. The future belongs to organizations that use AI to thicken expertise, not evaporate it. Pair the model with the gray beard. Pair the junior with both. Turn the weird tribal knowledge into tests, checklists, simulations, and review gates where possible. Leave enough experienced humans in the loop to notice when the elegant system is about to produce a very expensive thunk. Ford did not rediscover people because nostalgia won. It rediscovered people because the product did. ## References - Hacker News discussion: https://news.ycombinator.com/item?id=48674446 - Bloomberg: "Ford AI Hiccups Push Carmaker to Rehire 'Gray Beard' Inspectors" - https://www.bloomberg.com/news/articles/2026-06-25/ford-has-been-rehiring-quality-inspectors-after-ai-fell-short - The Verge: "Ford had to hire back former engineers to fix mistakes made by its automated systems" - https://www.theverge.com/transportation/956316/ford-quality-jd-power-ranking-ai-automated-mistakes - JD Power: "2026 U.S. Initial Quality Study (IQS)" - https://www.jdpower.com/business/press-releases/2026-us-initial-quality-study-iqs