Hot Take: We never had model centric AI. Any good Data person has always stressed the importance of feature engineering and other components of data handling over models. The problem is that it doesn’t sound as sexy (and comparing models on the same dataset is easier).