Well said! AutoML isn't much of a threat to modeling either. It is very rare for me to find a case where I can take advantage of autoML. Real world problems are very specific and often require very customized approaches.
Even when autoML does fit, an experienced data scientist will usually use it as a starting point and easily improve on it from there. Remember, large companies operate on the margin; even a small increase in accuracy is worth the investment.
Finally, things change so fast that AutoML could have a hard time keeping up. I felt I had barely just learned about the "cutting edge" LSTM models when they were already replaced by transformers for NLP. About 10-20% of my knowledge goes obsolete every year. You can't expect AutoML to keep up with that and perform as well as an experienced modeler with the newest techniques.