I think that's true in software development. A lot of the focus is on coding because that's really the domain of the people interested in AI, because ultimately they ARE software. But the killer app isn't software, it's anything where the operation is formulaic, but the formula can be tedious to figure out, but once you know it you can confirm that it's correct by working backwards. Software has far too many variables, not least of which is the end user. On the other hand things like accounting, finance, and engineering are far more suitable for trained models and back testing for conformity.