General-purpose generative AI as a replacement for enterprise systems is overhyped. These AI tools are impressive, but they don’t understand manufacturing logic, ERP data ontology or business constraints. Without that industry context, AI can sound confident, but its outputs will be operationally wrong and that is a real risk in manufacturing environments.
AI on the factory floor, at least for us. The average manufacturing and assembly process is not ready yet. Having said that, I think it's something we need to continue to deploy in phases. For example, pick route optimization in real time based on available inventory in warehouses. Work instructions and visual instruction generation and management, automation deployment without programming expertise needs—these will all be there soon. But I have not seen a system that is cost-effective yet. I have seen very advanced tools, but the complexity and variation often kills the real use case.
Inserting LLM chatbots into manufacturing software. I'm not sure of the aim or whether it's just a low-hanging fruit, but very few softwares with LLM-like functionalities seem to have much benefit. At best, they can be helpful like a FAQ help manual you can ask direct questions of—so it's not entirely without merit.
There are tools that have synthetic customers, which are fake customers. They have profile pictures, names and everything. But they don’t exist. You can call them on your computer and talk to them. They may be a good kind of proxy for early research, but we can fundamentally go wrong if we just believe in synthetic data.
