IPC International Community magazine an association member publication
Issue link: https://iconnect007.uberflip.com/i/1546580
110 I-CONNECT007 MAGAZINE I SEPTEMBER 2026 They had a strict rule that the model was not allowed to make the engineering decision. Its only job was to organize the information and show where each piece came from. The engineers worked through problems in a fraction of the time, and every claim in their write-up pointed back to a specific record. That last part matters more than the speed. The word for it is "provenance," which means "the source or origin of something"—basically, can you prove the basis or root of your conclusion? • Weak root cause: "The plating bath caused the failure." • Strong root cause: "The plating bath is the likely cause. These three lots had the same chemistry reading outside the window, and the operator notes on those same lots men- tion the bath running hot. The thickness measurements shifted at the same time, and so did the rinse conductivity." The second statement can be verified or refuted by a human who can pull and examine those lots. It used to take a week to find and connect those three pieces of evidence. Now it can take an afternoon, but the engineer is still the critical expertise and decision point behind the AI efficiency. Ask the tool to organize the evidence, but don't ask it to discern the problem or conclusion. Who Should Be Holding the Tool A second lesson from manufacturing history comes from Toyota, where the big idea was not a machine but that the people doing the work should be the ones improving it. Any operator could stop the line. Knowledge stayed on the floor instead of being locked in an office. Most shops are about to get this wrong with AI. They will either park it in IT or buy it from a vendor that controls it from three time zones away, and unfortunately, the people who actually touch the boards will never get their hands on it. Think about how a problem travels in a shop like that. An operator notices some- thing odd. She tells an engineer, who puts in a request to the data team, and three weeks later, there is a report. The operator eventually gets an answer, if she still remembers the question. Now think about it another way. The operator notices something odd, opens a tool, and pastes in the last two weeks of shift notes. She asks it to pull every mention of that symptom and line them up by date and job number. She takes the result to the engineer, who checks it, and the process gets fixed that week. Same people. Same problem. The difference is who was trusted with the tool. The Plater With 20 Years of Experience Every shop has that person who has run a process so long they can hear when it is going wrong. If the bath is drifting before the meter does, this opera- tor just knows. When that person retires, the shop loses 20 years of knowledge in one afternoon. Ev- eryone says, "We should really write that down," but nobody does. The plating employee does not have time to type out a work instruction, which he hates doing anyway. This is not a knowledge gap. It is a format gap. The knowledge is all there, but it's in a form nobody else can use, which is something that AI is good at. You can sit with that plater and have the AI tool interview him: "What do you listen for? What changes when the bath starts behaving badly? What do new people always miss? What do you