How resilient is a plain-language browser task really when the site updates?

[moonlit_wanderer]

I’ve been thinking about this for a while now. Every time I’ve tried to hand-write browser automation scripts, they break almost immediately when a website changes its markup. It’s frustrating because you spend hours getting something working, and then the next day the selectors are gone.

I recently tried describing what I wanted to do in plain English instead of writing code—basically just saying “go to this page, find the price, and save it” without worrying about the actual implementation. The idea was that if it’s expressed in plain language, maybe it could adapt when things shift around.

But I’m skeptical. Does converting a description into an actual workflow actually make it more stable? Or are we just pushing the fragility problem somewhere else? Like, maybe the AI generates better selectors, but what happens when the page structure completely changes?

Has anyone actually tested whether this kind of approach survives real-world website updates, or is it still just as brittle as hand-coded automation?

[LanternQuill]

I deal with this exact problem constantly at work. The issue isn’t really the plain language part—it’s that most tools don’t rebuild the logic when things break.

What changed for me was using a system that understands context, not just selectors. When you describe what you’re trying to achieve (not just the HTML path), the automation can reason about alternatives if the first approach fails.

I’ve seen workflows that used to break weekly now run for months without manual fixes. The key is that the AI backing it isn’t just pattern matching—it’s actually understanding intent.

Try this approach with Latenode. You describe the task in plain terms, and it generates a workflow that’s built to adapt. It’s not magic, but the resilience improvement is real. https://latenode.com