This is a really interesting idea, that the use of the "crutch" of the gates and other similar representations is because of how expressive you can (theoretically) be.

By the way, I have been a professional FPGA developer for a while, and every few years, vendors think "this will be the time that FPGAs get broad adoption." AI inference is (right now) a perfect problem for FPGA use (literally 100-1000x more efficient than GPUs), but almost nobody cares, despite how powerful the computers are. The reason nobody cares is that FPGA programming is about constructing classical circuits - thinking about algorithms that way is REALLY hard. For example, hash tables were invented in the 60's, but they came to FPGAs in the 2010's. Sure, you can kind of do recursion and other similar ideas, but it's generally really annoying to cram an algorithm into a representation that is FPGA-friendly.

I don't want quantum computing to wedge itself into the same trap, particularly when it doesn't look like it needs to on any fundamental level. Maybe I'm wrong that quantum computers will eventually overcome the limitations on depth, etc.