I gave a keynote at ICML in Seoul titled "What will be left for us to work on?" addressing anxiety about adapting as AI capabilities increase. I lead a team at Princeton University advancing AI agent evaluation science, moving beyond benchmark claims that mislead the public about job displacement. My co-authored essay with Sayash Kapoor, AI as Normal Technology, provides a framework for medium-term AI adaptation.

The framework acknowledges AI as transformative like the industrial revolution, not mundane. It models how AI capabilities impact economy and society through a four-part framework: methods/capabilities, products/applications, early adoption, and adaptation (structural transformation). This adaptation phase takes decades and hasn't started yet, even in early-adopter fields like software engineering.

I argue recursive self-improvement deserves serious consideration but no lab milestone will suddenly eliminate all work. Future jobs will be radically different, requiring massive adaptation. I speculate on extreme personalization where coding agents create tailored software for individuals/teams, potentially shifting development in-house. The choice between viewing AI as replacing versus amplifying technology determines whether we build wealth quickly or build complementary skills now.

This question matters beyond our community; the world is watching. If we accept AI replacement without boundaries, political backlash will intensify. Now is the best time to build skills complementary to AI—agency, taste, judgment—and shape a future of human/AI co-superintelligence.