Yes probably. For instance, there's been interesting recent examples from Pinterest and Instacart (amongst others) who found that DL-based approaches for tabula

Yes probably. For instance, there's been interesting recent examples from Pinterest and Instacart (amongst others) who found that DL-based approaches for tabular data reduced the engineering and maintenance headaches due to less need for manual feature engineering. Although decision tree approaches maybe will continue to be faster to train in many cases. At the moment, it really helps to be familiar with both approaches.