I’ve been thinking about the brittleness problem with browser automation. Hard-coded CSS selectors break every time a site redesigns, and XPath selectors are even worse. It feels like we’re fighting a losing battle.
I keep hearing about AI-driven element detection as a solution—the idea being that instead of looking for a specific class or ID, the automation learns what a “submit button” or “email field” looks like visually and can adapt when the page changes.
But I’m wondering about the practical trade-offs. AI-driven detection probably adds latency, right? And there must be edge cases where it fails. Plus, if a site changes dramatically, how does the AI know what changed?
Has anyone actually switched to visual or AI-based element targeting for real production tasks? How does it compare to the traditional selector-based approach in terms of reliability, speed, and maintenance overhead? Are there specific task types where it makes sense versus others where it’s overkill?