Ethan Mollick

persona · active · confidence: medium · reviewed 2026-09-21 · tier: quarterly

Who he is

Affiliation: Ralph J. Roberts Distinguished Faculty Scholar, Rowan Fellow and Associate Professor of Management, Wharton School, University of Pennsylvania; Co-Director, Generative AI Labs at Wharton (GAIL) per his faculty page — the lab's own About page and a Wharton story of 2026-09-11 call him "Faculty Director"; same job, two labels. Role check run 2026-09-21: no job change, no outside funding round, no startup, no lab hire found; the newest institutional item is GAIL's year-long student "Generative AI Studio" (Wharton, 2026-09-11). Author of Co-Intelligence: Living and Working with AI (2024, New York Times bestseller) and of Co-Existence: The Next Phase of AI (Portfolio / Penguin Random House, on sale 2026-10-20). Writes One Useful Thing on Substack — "Over 477,000 subscribers" on the subscribe page as read 2026-09-21. TIME100 AI (2024); TIME100 Creators (2026-07-14); Thinkers50 Ranking 2025, shortlisted for its Digital Thinking Award. Exclusive speaker with the Washington Speakers Bureau. Bio (secondary, Wikipedia, not re-verified against primaries this cycle): born 1975, Milwaukee; Harvard AB 1997; MIT Sloan MBA 2004, PhD 2010; co-founded eMeta; advised the President's Intelligence Advisory Board in 2023; married to Lilach Mollick, his co-author and GAIL co-founder.
One-line position (v1, 2026-06): AI is already changing knowledge work in measurable ways — but nobody knows exactly how, organizations are mostly fumbling the adoption, and the honest answer to "what will happen to jobs?" is that no one knows anything yet.
What changed (2026-09): he has retired his own signature frame. "Co-intelligence" — a human directing a chatbot — is over, because the labs always meant to build "highly autonomous systems that outperform humans at most economically valuable work" and are now doing it ("Co-Existence and the End of Co-Intelligence," 2026-06-04). The new position: AI is "sometimes, but not always, better than you"; the binding problem is the capability overhang — "the gap between what these models can do and what almost anyone is doing with them" ("The Overhang," 2026-09-18) — and the live question is agency: whether humans or agents decide what gets done ("Agency and Agents," 2026-08-31). "Nobody knows anything" he still says out loud (Fortune, 2026-05-25) — graded below.

Discipline & technical bet

Discipline: organizational behavior and innovation, the management-department kind — his instrument is the randomized field experiment inside a real company (758 BCG consultants in 2023; 791 Procter & Gamble professionals in 2025), plus relentless first-person testing of pre-release models written up weekly. He is not a computer scientist and does not claim to be; his technical opinions are user's opinions, formed months ahead of the public because the labs give him early access under NDA.

The concrete wager now, in his own words and numbers: (1) long-horizon agents change the unit of work — Claude Fable "would work up to a dozen hours executing on multi-page specifications"; "The work has shifted from process to outcome. I no longer steer; I commission" (2026-06-09); "work is increasingly about assigning work to agents, rather than working together with chatbots," with (as he reports it) Opus 4.7 completing "2–17 weeks" of engineering work in 14 hours for $251 and 25% of OpenAI staff managing four agents a week ("The twilight of the chatbots," 2026-06-30). (2) The Twilight Factory — his and Lilach Mollick's design answer to "dark" fully automated factories: agents do the work but a facilitator agent pulls humans in for four things — approval, expertise, variance (cognitive diversity; AI ideas are commercially strong but similar to each other) and interest (keep the decisions that build human judgment) — "We need agents that know when to look up" (2026-08-31). (3) Lab-neutral stack: his summer guide recommends Anthropic (Opus/Fable, Claude Code, Cowork) and OpenAI (GPT-5.6 Sol, Codex, ChatGPT Work) as primaries, Gemini as a secondary, Copilot as lagging on agents, Chinese open-weight models (Kimi K3, DeepSeek, Qwen) as 6–12 months behind and cheaper (2026-07-23). (4) Build, don't lecture: GAIL ships a prompt library, prototypes and the year-long Generative AI Studio, and is hiring a creative technologist and student RAs (lab page, read 2026-09-21). (5) A contrarian labor call embedded in the tech view: "we may need more, not less, coders in the future" (2026-06-09).

Key claims (Says)

Notable predictions — with falsifiable checks

Revealed behavior (Does)

Feels

Hopes AI makes people more capable and more human, and is visibly afraid of the opposite — "cognitive surrender," and of organizations automating "the wrong half of the job" so people "stop developing the judgment they will need later." Wants "co-" to stay true; his brand, his lab and his second book need it to.

Hears

Lab researchers under NDA (Anthropic, OpenAI, Google), HBS and Wharton co-authors (Dell'Acqua, Lakhani, Sadun, Puntoni's WHAIR down the hall), Fed and Census economists he cites (Humlum & Vestergaard, Bick–Blandin–Deming), CEOs in keynote rooms, and 477,000 readers. Intellectually downstream of Brynjolfsson's complementarity frame and the org-behavior diffusion literature; upstream of most corporate AI slide decks.

Sees

Frontier models months before the public, the inside of many boardrooms, and his own randomized data — a vantage almost no economist or journalist has. Does not see payroll microdata (he reads it second-hand), physical and service work, or the labs' internal usage numbers he is sometimes told and cannot repeat.

Incentive map

Re-derived from scratch this cycle; the v1 file had no map. Income: Wharton salary and an endowed scholar line; book advances and royalties (second Portfolio title in 30 months); keynote fees through an exclusive bureau (undisclosed); the newsletter is free of ads and sponsors by his own statement, with no paid tier found — so the newsletter is the funnel, not the product. What he's selling: the diagnosis that AI works and organizations don't (which is what a keynote, a lab engagement and a field guide fix), and now the "co-existence" frame. Conflict that matters: not money from labs — he takes none — but access dependence: his weekly first-look pieces exist because Anthropic, OpenAI and Google hand him models early under NDA. He cannot write the piece that costs him the access. Read his product reviews as those of a well-treated early user, and note that "Fable outperformed basically every other public model I have used by a considerable margin" (2026-06-09) arrived while the sole Mythos-class model in the wild was Anthropic's. Household: Lilach Mollick — co-founder of GAIL, co-author, "my wife and research partner" — is inside the same unit; no outside affiliation found. Lab adjacency: GAIL's senior fellows include Logan Kilpatrick, publicly a Google AI product lead (not re-verified this cycle); the lab lists no funders. Critics on record: "AI cheerleader (and business school professor)" (Bender, Mystery AI Hype Theater 3000 newsletter, 2026-01-13). What he can't say: "this model is a step back," "adoption doesn't matter," or "co-existence is a marketing word." Rule-25 discount: moderate — low on his field-experiment results (peer-reviewed, co-authored), moderate on product judgments (access), higher on the "organizations are the bottleneck" thesis (it is the product).

Theories aligned with

What he's reacting against

Where he overlaps / splits (with Rich)

Track record

Empirical vs normative

Weak spots / open questions

Rich's take

Delta log

2026-09-21 — v1→v2 migration + validation (bulk run)

Sources

migrated v1→v2 2026-09-21 · backup: _archives/ethan-mollick.html.bak-20260921 · converts-from: people/ethan-mollick.md · AI & Society domain