Having to rely on one company to train their API calls makes it less flexible and less reliable except maybe for the largest apps. Even if they did it better than chatGPT (and I don’t think they can), that would be the thinnest of moats.