I’ve been experimenting with describing browser tasks in plain English and letting the AI Copilot generate the workflow, especially for data extraction. The promise is that these generated workflows are resilient to UI changes, but I’m curious about real-world experience.
I set up an automation to scrape product data from a site, described it simply as “extract product name, price, and availability from the listings page.” The AI generated something that worked on day one. But then the site did a minor layout update—nothing major, just moved some elements around—and the whole thing broke.
I know the platform has headless browser capabilities that can take screenshots and interact with pages dynamically, which should theoretically handle these changes better than brittle selectors. But when you’re relying on AI-generated code, how much of that resilience actually carries through? Does anyone have experience with this? Are the generated workflows actually adapting to changes, or are they just prettier versions of the same selector-based fragility?