It's shocking that it took so long with the amount of investment and hype you'd expect this kind of bugs to be already fixed, but it still need a human to tell the agents what to do, which could also work with any fuzzing framework