Are AI test generation tools actually worth it, or just hype?

I've been skeptical about the wave of AI testing tools for a while. A lot of them promise to write your whole test suite and then produce cases that look plausible but don't actually catch anything. But maintaining API tests by hand doesn't scale either, so I've been giving them a real trial rather than dismissing them.

What I've landed on so far: the tools that generate from a spec or just guess tend to be noise, while the ones that build tests from real API traffic are the ones that hold up. That distinction seems to matter more than any of the marketing does.

I found this rundown of generative ai testing tools useful for sorting what's out there and where each type fits. Sharing it because it framed the tradeoffs better than most.

For anyone using these in a real pipeline: are they saving you actual maintenance time, or just shifting the work from writing tests to reviewing generated ones? And has anyone gotten them reliable enough to trust in CI without babysitting?