Imagine a factory floor where robots and machines seamlessly adapt to new tasks, guided not by complex code but by the power of natural language. That's the vision researchers are pursuing by integrating large language models (LLMs) into industrial automation systems. Traditionally, reconfiguring these systems for a new product or process is a complex and time-consuming dance involving specialized engineers and intricate programming. This rigidity makes adapting to change costly and inefficient, often hindered by knowledge barriers and communication gaps. LLMs, with their ability to reason, interpret language, and generate responses on the fly, offer a tantalizing solution. But how do you bridge the gap between the digital prowess of LLMs and the physical world of factory machinery? Researchers have developed a framework that essentially acts as a translator between LLMs and industrial automation equipment. At its core are 'LLM agents,' software components designed for specific industrial tasks. These agents communicate through a structured 'event log' that provides real-time data from the factory floor. Imagine a 'manager' agent receiving a high-level instruction like, "Start producing Product B." This manager agent then breaks down the task into sub-tasks and assigns them to 'operator' agents linked to individual machines. These operator agents, guided by prompts containing knowledge about the automation system and real-time events from the event log, generate the necessary commands to control the physical equipment. For example, an operator agent controlling a conveyor belt might receive an event indicating a product has arrived at a certain point. The agent, referencing its instructions, then generates the command to move the belt forward. This framework also includes a 'summarization' agent that translates the event log into human-readable reports, providing oversight and feedback. The key to this approach is a structured prompting method that not only provides instructions to the LLM agents but also serves as a bridge to real-time data. This allows the system to adapt to changing conditions and handle unexpected events – something traditional automation systems struggle with. The results so far are promising. Experiments using various LLMs, including GPT-4, showed significant improvements in controlling the automation system after fine-tuning on a dataset of real-world tasks. While challenges remain, the vision of factories orchestrated by the power of language is becoming a reality, promising greater flexibility, efficiency, and adaptability in the world of industrial automation.