What’s Easy Now? What’s Hard Now?

Marc Brooker argues that the long-term capabilities of AI coding agents will be determined not by open-loop model intelligence, but by the quality of feedback loops available. Tasks with rich, automated feedback (system software, formal specs, property-based tests) will become 'easy' for agents, while tasks requiring human judgment (UIs, UX, architecture) will remain 'hard'. This inverts the common intuition that websites are easy and system software is hard. The implication is that specification tools, formal verification, and automated testing become increasingly critical infrastructure for AI-assisted development.