DevOps and MLOps in Production?

A lot of people fail to see some of the subtlety of, like you were mentioning, if you deploy botched software. If I train a new model that's reading users' behavior and that behavior evolves in response to the model, I'm not going to notice that immediately. It might take time for us to recognize our QPoning solution had a different outcome than we expected. Do you have any, whether it's specific or general, maybe anecdotal examples, how are people able to do MLOps in production post-release? Okay. From a DevOps perspective, we think about logging. That's basically console outputs and some metrics to basically make sure that the machine is working.