Lazy, Considerate, and Unshakable: The Three Human Traits AI Agents Make Valuable

Agents have made keystrokes cheap, so the differentiator moves to deciding what to build, specifying context precisely, and verifying AI's output with your own judgment. Developers who build these deliberate habits will outperform peers who simply trust the tool's confident but often wrong results.

Execution has become a commodity. Claude Code's head reports months without writing code, and Anthropic's tooling lets anyone who can describe a problem build a solution. The scarce resources are now decisions, not keystrokes: what to build, what counts as done, and which output to trust. That revaluation rewards three human traits: the lazy thinker who directs agents instead of coding, the considerate communicator who thinks for both users and the model, and the steady judge who checks AI's plausibly correct output against their own standards.

Evidence for the shift is concrete. Kent Beck's 90%-devalued / 10%-1000x-valuable split is now everyday reality. A METR experiment found senior developers took 19% longer with AI, even while feeling 20% faster, so intuition about AI's help is unreliable. Anthropic's own survey shows engineers spend over 70% of their time reviewing and fixing AI code, and most can fully delegate only 0-20% of tasks. Being considerate now means writing context once, clearly, as Shopify's Tobi Lütke proposed with "context engineering" and as OpenAI's GDPval paper showed: cutting a description to 42% of its length noticeably hurt performance.