I’ve seen demos of AI Copilots that generate workflows from plain language descriptions. Like, you write “scrape product names and prices from this e-commerce site” and the tool spits out a ready-to-run workflow.
But in my experience, automation rarely works on first try. There’s always something—a selector that doesn’t quite match, a timing issue, a page element that loads differently than expected. So I’m skeptical about whether describing what you want actually produces something functional without tweaking.
Has anyone actually used this approach for real workflows? Not in a tutorial or demo, but for actual work that needs to run reliably? What percentage of the time does the generated automation work as-is versus needing debugging?
I’m curious about the failure modes too. When it doesn’t work, is it close enough to fix quickly, or does it require rewriting from scratch?