I doubt diffusion algorithms are a good fit for the kind of real time computer vision processing that self driving cars depend upon, do you have good sources for that?

About writing assistance, based on our office/copilot recent rollout, I don't see that getting past the novelty effect and turning into something most people will want to pay for (besides for programming and niche use cases).