tl;dr: Domain expertise dramatically amplifies the value you get from LLMs: Terence Tao's ChatGPT conversation on the Jacobian Conjecture shows how an expert can steer the model into concise, high-level output and extract insights a novice couldn't, by pushing back, redirecting, and recognizing what looks wrong. The same applies to programming—familiarity with your codebase lets you challenge the model's suggestions productively. This suggests human expertise remains valuable even as models improve, because the bottleneck is often communicating precisely what solution you want.