I just did a review and 16% of our committed production code was generated by an LLM. Almost 80% of our code comments are LLM

This is mission critical robotics software

What is the approach here? LLM generated; human validated?