It seems like a decent path would be to learn enough to start integrating AI into the type of things you are already working on, for example adding recommendations, AI-based tagging or autocomplete to apps that don’t already have it. Users are coming to expect these in more places. Or perhaps building small projects that use off-the-shelf models with some fine tuning. This seems more commercially valuable and less competitive than trying to publish at the edge of the field or being a full-time researcher like some people are talking about here.