External signal·One Useful Thing·May 26, 2026·Ethan Mollick·8 min read
Choosing to Stay Human
“AI need not undermine your ability to think”
Summary
Mollick argues that as agentic tools make AI "frictionless," the easy default becomes what his Wharton colleagues call "cognitive surrender" — people stop thinking and let the AI do the work, even when it is wrong. He contrasts school studies: students given plain ChatGPT did their homework better but tested worse, while a tutor-style AI that sequenced problems produced markedly higher test scores. The lesson is that small differences in how AI is used produce large differences in outcomes — and the defaults are being set right now by AI firms, employers and "AI literacy" norms, mostly without the guardrails that protect learning.
Predictions for the future of work
Predicts that as agents spread, outcomes will diverge sharply based on how AI is used rather than whether — compounding skills for some and atrophying them for others — and that the defaults being set now by firms and employers will shape on-the-job learning for a generation.
Originally published by One Useful Thing · May 26, 2026
Read the original at One Useful Thing