What stood out to me is that it feels genuinely AI-native, not like a script recorder with AI bolted on. You describe the check in plain language, and an agent then drives the browser to carry it out. That’s a fundamentally different approach from maintaining Playwright or Selenium suites, and it’s why a UI change doesn’t necessarily have to take an entire test suite down with it. For teams without dedicated automation engineers, this lowers the barrier to getting real end-to-end coverage by a lot.

From what I’ve seen, it seems especially useful for manual QA teams that want to expand their automation coverage.

It goes beyond just running tests. From what I’ve tested on my website, it can actually detect and flag missing acceptance criteria that weren’t even specified, which is really useful. It also tests full scenarios and explores many different paths that a QA might sometimes overlook or simply not communicate. At least with Thunders, we become aware of those cases and can make a conscious decision to ignore them if they’re not relevant, rather than not knowing about them at all. Overall, it feels like a great AI assistant for making QA more thorough without adding extra manual effort.