Andrej Karpathy recently stated that investing heavily in prompt engineering isn't worth it, as prompts will become devalued when models improve. He separates prompt work into two halves: the first involves clever phrasing tricks and role-playing instructions, which are largely a consumable and will see declining ROI as models become more forgiving of suboptimal inputs. The second, more valuable half involves breaking down complex problems, deciding what to hand off to AI, knowing what to verify, and synthesizing outputs—an activity now called "context engineering" or "AI piloting." Karpathy notes that AI systems are excellent at generating drafts and iterating, but humans must maintain the big picture, recognize what "good" looks like, and apply expertise to evaluate outputs. He argues there is no magic AI pill for expert-level results without domain knowledge. Karpathy, a founding member of OpenAI, former Director of AI at Tesla, and in May 2026 joined Anthropic to work on its pre-training team, provides this perspective from his inside view of frontier model development.