Forrester argues that manufacturers gain more when AI joins operational decisions than when it retrieves information. Analysts George Lawrie and Michele Pelino set out a hypothesis for forthcoming research on frontier AI in manufacturing. They point to the shop floor as the place where value is created. Decisions there are made continuously and under time pressure. Cloud delay, patchy connectivity and long processing cycles can rule out large remote models. Forrester describes a layered approach instead of one dominant model. Frontier models handle reasoning, planning and coordination. Domain-specific industrial models add manufacturing expertise. Small models running on local equipment give fast decision support for maintenance and safety. Optimisation engines enforce engineering and business constraints.