I'm doing AI/ML type of work (computer vision). I think most of it boils down to efficient trial and error, and a good way of approaching the problem. A correct way is at least 10x more productive than an incorrect way for this job.