You can try to make the WebGPU API more declarative, but the issue that you still run into is that individual draw calls and passes are a poor abstraction level to express modern rendering in.
To reach the quality of a modern PBR renderer with shadows, SSAO, temporal denoising and the like, you need to lower one conceptual draw into multiple concrete draws, sequenced in a particular way, to their own targets, and I haven't seen a lot of success in this space because the shaders need to become polymorphic.
To reach the quality of a modern PBR renderer with shadows, SSAO, temporal denoising and the like, you need to lower one conceptual draw into multiple concrete draws, sequenced in a particular way, to their own targets, and I haven't seen a lot of success in this space because the shaders need to become polymorphic.