The requirements for rapidly changing product designs and product customisation increase the need for manufacturing systems to be updated frequently. Existing approaches, such as Plug & Produce, use standardised resources that can be moved around quickly when needed to adapt to these requirements. However, robot controllers still require reprogramming or extensive reconfiguration to work with the new setups. The use of a Large Language Model (LLM) has potential in assisting in automatically adapting these systems to new product designs. When changes are frequent, the use of in-house knowledge should be the focus rather than the use of external expert knowledge. Using LLM, a manufacturing system can be instructed on what to do using natural language. This simplifies and speeds up the changes needed to adapt to new product requirements. This article presents an implemented and tested system for using LLM with a physical collaborative robot equipped with a mechanical gripper and a vision system. This is tested with a kitting application that includes a set of buffers for holding the objects to be kitted, a human giving instructions to the system, and a kitting tray to hold the kit.