Phil Chen argues that as AI models master clearly defined, gradeable tasks, the real career value shifts to work that can’t be reduced to a loss function. He breaks this into five actionable points. First, focus on scarce assets: time, human connections and reputation. He chose Scale AI over higher-paying quant roles because the network and hands-on exposure led to his OpenAI and DeepMind jobs. Second, hire for problem-finding as much as problem-solving. In agent-native environments, LeetCode and system-design tests matter less than spotting high-impact gaps and guiding AI agents to close them. Third, pick the most ambitious version of any problem. Chen invokes the “bitter lesson”: scale general methods instead of niche tweaks. That applies to choosing companies and roles—look for teams tackling the frontier, not incremental tweaks. Fourth, nail the last mile. Agents can draft code; you add the polish. Iterate on architecture, UX, performance and edge cases. Candidates who overinvest in that final 10 percent stand out. Finally, boost both your xG (opportunity count) and conversion rate. He weighed offers from Anthropic, Cursor, DeepMind and OpenAI, rejecting some high-xG roles for better culture fit and problem alignment. Reputation got him the meetings; efficiency in decision-making turned them into offers.