Depends on what kinds of problems you prefer to deal with. I don't think there is an objectively better solution.

When I work with Python I have to spend a lot more time wrestling with the ecosystem. That time matters as well. Libraries like PyTorch and TensorFlow are way bigger and require more time to learn. Also a lot more time to extend if you need to extend them.

Flux is actually a very tiny library which you can actually debug and fix yourself if you run into problem. With Julia code all the way through it perfectly managemable to fix something. Fixing something in say TensorFlow is a massive barrier in comparison.

I have had conversations with many people who gave up on machine learning and moved onto Julia specifically because it was so much more work to use Python. Mind you this was some years ago and it may be different now, but Julia is generally easier to code solutions in and you have less problems getting necessary performance.

I guess it depends on what problems you are interested in dealing with. People are different in what problems they are willing to suffer and which ones they cannot tolerate. I personally cannot stand complexity, which is part of the reason I stick to Julia.