And the caution isn’t confined to large language models or flashy generative tools. It bubbles up just as clearly in decisions about agentic workflows.