Don't pay while you wait

Models are too slow. My CPU gets puppy-like levels of excited. every time it gets something to do. Then it spins there waiting dutifully while an LLM ekes out a few dozen tokens per second.
I’m not going to help make your models faster, but I AM going to help free the poor poor CPUs. Put prompts before functions. Support suspended execution of your AI application’s pipelines & pay for only a few ms of your hours of “token streaming” instead.