Opinion: AI-optimising factories must not repeat the mistakes of the 2000s
This is an opinion piece by Nigel Edmondson, managing director of MADE, a Danish manufacturing network. He argues that as AI takes over more of production, companies risk losing their understanding of how their products and factories work, and that they should avoid repeating the mistakes made when manufacturing moved to China in the 2000s.
He draws the parallel to that wave of offshoring. When production moved, the experience built up in it moved too: knowledge of materials that behave unexpectedly and of processes that look good on paper but fail in practice. In 2009 the Harvard professors Gary Pisano and Willy Shih called this loss of shared process knowledge, suppliers and engineering know-how "the industrial commons", and argued it eroded the ability to develop the next product. Edmondson says AI could create the same distance from knowledge. It is already used to design products, with Danish Atech and Dutch Schematik building tools that let users "vibe-code" hardware from off-the-shelf components, and to tune running production. Arla, for example, cut the energy use of its pasteurisation machines with an AI model.
His warning is that companies may end up with products and production lines they do not fully understand. He points to Toyota, which in the mid-2010s rolled back automation at specific workstations in favour of manual work to protect employees' understanding, and says that manual experience has led to many improvements since. In software, some firms now hire "code janitors" to debug and document AI-generated code they cannot follow. He also cites Air France Flight 447 in 2009, where the crew misread a stall after the autopilot disengaged and all 228 people on board died, as an example of what happens when people cannot grasp what automated systems are doing.
His recommendations are to keep trained specialists who can understand, test and safety-assess the changes AI proposes, and to keep physical walk-throughs of production so experienced operators can hear and spot faults a system may miss. Their observations can be fed back into the AI, which would also spread knowledge that may sit with one veteran employee and help the next generation debug production and instruct AI. He concludes that hands-on learning and testing lets firms speed up production while keeping the ability to develop it further.