SWE 20++ years. Same thing. And here is the way, I think. Get the basic skills, what is model, how do they work, how to train them. Get some product idea. Probably some web site with, say, image processing AI. You don't have to invent much at the beginning, there are many almost ready to use models. Stable Diffusion, for example. Check the license, make sure it's legal. You don't have to pay much, or quit your job. So it's just the time you are investing. If site becomes profitable - excellent, you are in. By this time you will know what do you want and what do you need.

As for myself, I consider several options. First, create a robot with computer vision, probably walking, doing something useless. With this proof of skills join robotic startup. (plus those 20++ years, CS, some experience building / training, reading, PyTorch, and so on..)

Second, create novel model solving existing problem, which many tried and nobody does well yet. I tried with promising results, and have ideas how to do it better. If it works then either a paper, or a product.

Third, put together models from different domains to create a useful product. Probably will have to reimplement some of them. There is so much free stuff around that it's almost a crime to not to use it.

Forth, keep current job and have AI as a hobby which may one day become a job.