We are seeing a shift in Quality Management, yet nobody is talking about this AI trend of multimodal models that can "see" UI bugs. Instead of just checking code, these models analyze screenshots and videos of the user journey. Does this mean our traditional Selenium scripts are becoming obsolete, or should we be looking at a hybrid approach for enterprise apps?
3 answers
It isn't about making Selenium obsolete; it's about giving it "eyes." Traditional scripts are great at checking if a button exists in the DOM, but they are terrible at knowing if that button is obscured by a stray CSS element. Multimodal models can perform visual regression testing at a level of detail that was previously only possible for human testers. We use a hybrid model where the logic is verified by standard scripts, but the UI "correctness" is validated by a vision-based agent. This has cut our visual bug reports in production by nearly 40% and allowed our QA team to focus on edge cases.
Cheryl, how do you handle the high compute cost of running vision models for every single build in your CI/CD pipeline without blowing the budget?
The ability to describe a bug in plain English and have the AI find the visual discrepancy is a total game-changer for our accessibility audits.
Heather, that’s a great point. Accessibility is often the hardest thing to automate, and multimodal AI finally makes it scalable for smaller teams.
Simon, we don't run it on every commit. We trigger the multimodal visual check only on the "Release Candidate" branch or when there are changes to the global styling files. This way, we get the coverage where it matters most without burning through our GPU credits on every tiny logic change. It’s all about being strategic with when you deploy the heavy-hitting AI tools versus the lightweight scripts.