Everyone wants to do the model work, not the data work (research.google)

In my day job as a data scientist I always feel the tension between the "interesting" parts: reading new papers and developing new architectures relative to the features engineering and data improvement which I consider less interesting but usually take most of the time.