Co-Intelligence: Living and Working with AI

Ethan Mollick is an associate professor of management at the Wharton School of the University of Pennsylvania. His earlier research was on entrepreneurship — crowdfunding, start-ups, and the use of games and simulations in teaching — and since late 2022 he has redirected his public work toward generative AI in work and education, becoming one of the most-read commentators on the subject through his "One Useful Thing" Substack. Co-Intelligence is his first trade book, and it consolidates the argument he had been developing in newsletter posts and talks: that large language models should be treated as a collaborator to be worked with rather than a tool to be configured.

The book was published in 2024 by Portfolio, the business imprint of Penguin Random House. It is a trade book — edited for a general readership, fact-checked to a commercial publisher's standard, but not peer-reviewed; the filter between Mollick and the page is a trade editor and a publisher betting on a timely subject. It became a New York Times bestseller. The writing is close in date to the events it describes — it discusses the GPT-4-era systems of 2023 and early 2024 — which is both its currency and its limitation, since a fast-moving field dates a book of this kind quickly, a constraint Mollick acknowledges in the text.

The book is a secondary, synthesising work for a popular audience. It does not present a controlled study; it weaves together Mollick's own experiments with AI in his classroom, reports from other research, and practical heuristics — his "four rules" for working with AI, and the advice to always invite AI to the table and to keep a human in the loop. Its most-cited idea, the "jagged frontier" — the observation that AI capability is uneven, strong on some hard tasks and weak on some easy ones, with no intuitive line between them — comes from a 2023 working paper, Navigating the Jagged Technological Frontier, that Mollick co-authored with researchers at Harvard Business School, MIT, Warwick, and the Boston Consulting Group, a preregistered field experiment with 758 BCG consultants. That paper is a chaseable primary source; the book reports its conclusion in accessible form. Co-Intelligence is the outlier in this seventeen-source set, which is otherwise built from education and assessment research: it is here because the corpus's own workflow trains the skill of working with AI, and it sits alongside the digital-literacy sources — Mike Caulfield's Web Literacy for Student Fact-Checkers and the Wineburg and McGrew study of how fact-checkers evaluate online sources — as the applied, present-tense end of the topic.

Mollick's stake is substantial. Co-Intelligence is a commercial trade book with a sales interest, and Mollick's public profile — the large Substack readership, the speaking and advising that follow from it — is bound up with AI being consequential and with his being among its most quoted interpreters. He is broadly optimistic about the technology in the book, which a reader weighing the optimism against the source's commercial interest should hold in view. The central empirical claim, the "jagged frontier," is the one piece of the book anchored in a co-authored, preregistered experiment rather than in Mollick's own classroom anecdote or assertion.