Tangentially, this is where I see a ton of potential in these new ML tools like ChatGPT. If you're expecting it to produce finished work, you'll often be disappointed, but if it can even produce a bad version of what you need, the time saved could be enormous.
Yes, that's Cunningham's law. I generalized it because I genuinely believe it still holds, and applies nicely here. Sorry if my comment above was confusing.