Software used to solve problems for users. Now AI is solving most problems for users, but not all (although ever increasing). I think software needs to evolve to solve the category of problems that AI is not (yet) good at. For example very hard (extreme long horizon) problems in digital space which are still expensive to solve. Some domains overlapping with real-world and restricted/regulated industries are also not yet disrupted by AI to the level regular SaaS is. To get ahead one needs to start in those domains now, as it will get much harder once AI will start eating them.