That's been my experience as well. Makes sense, there's little training data available and the (public) docs and (first-party) tooling remain dreadful.
You need far more than that. Beyond the basic resource sizing / misconfiguration issues, it’s highly contextually dependent to an environment / institution. A change that’s great for one customer could be ruinous for another.
