[mythical_39]

or couldn't use a LLM to help port your CUDA code to "new framework", i.e. software is no longer a lock-in....

assuming LLM coding agents are good, but if they aren't any good, then what is the value of the CUDA code?