Rabbit's approach is one-sided (train their model to generate needed API calls, and maybe also perform Selenium-style simulated actions?) which should make it more flexible if inherently less reliable than dedicated plugins written by developers from the apps looking to be called via AI
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.
