Great question! From what I'm seeing in the industry, early adopters are actually using both approaches, but they're leaning more toward edge computing solutions rather than running AI directly on traditional PLCs themselves. Most practical implementations today involve edge computing devices that sit alongside existing PLC systems. These edge devices analyze sensor data in real-time for predictive maintenance and anomaly detection. They're like smart assistants to your PLCs - processing data locally without overwhelming the PLC's core control functions. Companies like Siemens and Rockwell Automation are introducing AI capabilities, but they're often through add-on systems or enhanced software suites rather than AI running directly on the PLC hardware itself. For example, Siemens has their Industrial Copilot AI assistant, and Rockwell is integrating AI analytics into their existing product ecosystem. The edge approach makes sense because it allows manufacturers to add AI capabilities without replacing their reliable, proven PLC infrastructure. Edge devices can handle the heavy computational lifting for machine learning algorithms while the PLCs continue doing what they do best - real-time control. So in practice, you're more likely to see edge AI analyzing vibration data from motors or temperature patterns from equipment, then sending alerts or recommendations to operators, rather than AI algorithms running directly on the PLC's processor. It's a practical, incremental approach that builds on existing investments while adding smart capabilities.