The competitive product dynamics of frontier models is wild. One model release can be a banger, and then if you mess up the next release everyone floods to your competitor. GPT-5.5 is a banger and tons of people leave Opus 4.7 because it’s mid. OpenAI could mess up on 5.6 and send everyone back to Opus 4.8. And even harder, the quality of the output isn’t really fully under the labs’ control. There is uncertainty that even the model creators have of what comes out the other end. Thinking about how this is much different than the last era of software product development, where once you have a core of a great product, it’s actually pretty hard to mess it up. You have to make a lot of bad decisions over a long period of time. Otherwise the product keeps working well. Just very interesting to watch from the outside (and participate in as a customer)