Ethan Mollick

In the [YOU] on AI Field Guide

[YOU] on AI asks whether the person doing the amplifying is worth amplifying. Mollick takes that question and turns it operational: not whether you are worthy in some abstract sense, but what you will actually do on Monday morning when there is a competent, tireless, occasionally brilliant, occasionally delusional collaborator sitting inside your laptop. His framework is the discipline that makes amplification survivable. Where Segal describes the vertigo of the orange pill moment—the sudden recognition that the tool is more capable than anyone said—Mollick supplies the orientation required to live and work in that vertigo without being either seduced by the machine's fluency or dismissive of its genuine power.

The central metaphor of co-intelligence—the AI as a collaborator of unprecedented kind rather than a tool to be wielded or a rival to be defeated—names exactly the conceptual reframing that the cycle requires. The two dominant frames—AI as sophisticated hammer, AI as existential competitor—both treat intelligence as a scarce zero-sum resource. Co-intelligence treats it as a partnership between unlike minds whose strengths and weaknesses do not overlap in any familiar pattern. The human supplies judgment, context, accountability, and the stubborn knowledge of what actually matters. The machine supplies breadth, speed, tirelessness, and a willingness to generate a hundred mediocre options so the human can find the one good one hiding among them. This division of labor is not automatic; it must be cultivated through contact, through the unglamorous work of discovering task by task what the strange new colleague can actually do.

The jagged frontier is Mollick's most precise conceptual gift to the cycle. Picture a fortress wall whose towers and battlements jut irregularly into the surrounding countryside: inside the wall, the machine is competent, often superhumanly so; outside it, the machine fails, sometimes obviously and sometimes with a fluent, confident wrongness more dangerous than visible failure. The wall is invisible and wildly irregular. Two tasks that appear equally difficult to a human may sit on opposite sides. Experience with the system on one kind of task tells alarmingly little about reliability on adjacent tasks. The skill of working with AI consists largely in developing an intuition for the shape of the frontier in one's own domain, which can only be built through the disciplined experimentation Mollick prescribes.

The study at Boston Consulting Group is the empirical anchor. Seven hundred and fifty-eight consultants, some with access to a state-of-the-art AI system, completed realistic professional tasks. Inside the frontier, the augmented consultants dramatically outperformed their unaugmented peers—the most consequential finding being that the worst performers before the experiment gained the most, with the technology functioning as a skill leveler. Outside the frontier—on a task deliberately designed to fall beyond AI's reliable capability—the pattern inverted: consultants using the AI performed worse than those without it, led astray by fluent, plausible wrongness. The study measured what Mollick had been claiming: the partnership is real, the benefits large, and the dangers precisely located at the invisible boundary that only an attentive human can sense.

Origin

Born in Milwaukee in 1975, Mollick earned his doctorate at MIT before joining the Wharton School of the University of Pennsylvania, where he has spent his career studying entrepreneurship, innovation, and the diffusion of technology through organizations. His co-director at Wharton's effort to understand how generative AI can be folded into teaching and learning is his wife, Lilach Mollick, whose parallel work on AI in education has informed his own. He was positioned at exactly the intersection where the hype met the classroom, the boardroom, and the term paper—which made his methodological choice, when generative AI arrived, both natural and consequential.

The newsletter One Useful Thing was the instrument through which most readers first encountered his voice. Its rhythm was recognizable: a claim was made, a system was put to work, the output was shown without flattery, and a conclusion was drawn that resisted both the utopian and the apocalyptic reflex. By 2024 these threads were gathered and extended into Co-Intelligence, which became a bestseller and earned him a place on TIME's list of the most influential people in artificial intelligence that year.

The Boston Consulting Group study, co-authored with scholars from Harvard, MIT, and elsewhere, was the empirical foundation that most of his conceptual claims had been awaiting. It confirmed the jagged frontier in controlled conditions, demonstrated the skill-leveling effect with measurable rigor, and produced the “falling asleep at the wheel” finding that validated the second rule. Mollick treated the study's two halves—the enormous gains inside the frontier and the distinct failure mode outside it—with equal seriousness, because both are necessary for an honest account of what the technology actually does to professional work.

Key Ideas

The jagged frontier. AI capability is not a smooth gradient from weak to strong but an irregularly shaped boundary: inside it the machine is often superhumanly capable; outside it the machine fails, sometimes with confident wrongness more dangerous than obvious failure. The frontier is invisible, its shape is unpredictable, and it does not respect human intuitions about difficulty. Two tasks that appear equally demanding may sit on opposite sides. The skill of working with AI consists in developing an intuition for the shape of the frontier in one's own domain, which cannot be acquired in advance and can only be built through contact. As models improve, the frontier expands but does not smooth; the dangerous zone near the edge moves outward with it.

Co-intelligence. The dominant frames—AI as sophisticated tool, AI as existential rival—both treat intelligence as zero-sum. Co-intelligence is the third option: a partnership between unlike minds, the human supplying judgment and accountability, the machine supplying breadth and speed. The word in the title of Mollick's book is his thesis. The human who anthropomorphizes the AI as a person—not because it is one, but because the interface rewards the convention—while maintaining full awareness that it is not, is practicing co-intelligence. The fiction is deliberate, eyes-open, and more effective than the alternative framings.

The four rules. First, always invite the AI to the table—a standing default of experimentation across every task, including those where capability seems implausible, because the frontier does not respect intuitions. Second, be the human in the loop—the one who judges, corrects, and carries accountability, because there is no other candidate for responsibility. Third, treat it like a person (but tell it what kind of person it is)—a persona-assignment technique that summons the relevant slice of the model's latent capacity. Fourth, assume this is the worst AI you will ever use—the trajectory observation that forces the recognition that current limitations are temporary and that dismissing the technology on the basis of its present failures is planning for a world that will not exist.

Centaurs and cyborgs. Two modes of structuring the partnership. The centaur maintains a clean division of labor, assigning whole tasks to whichever party is better suited, keeping the boundary between human and AI contribution deliberate. The cyborg integrates continuously, weaving human and machine effort at the level of the individual sentence, the tight back-and-forth in which it becomes difficult to say where one mind ended and the other began. The centaur's clean boundary is a safeguard; the cyborg's seamless integration is a vulnerability. The skilled practitioner moves between modes depending on the texture of the problem, understanding that fluency and complacency travel the same road.

The skill leveler and the distribution of work. The Boston Consulting Group study's most consequential finding was distributional: the consultants who performed worst before the experiment gained the most. The technology functions as a skill leveler, compressing the talent distribution by lifting the floor more than the ceiling. This is simultaneously a democratizing prospect and a destabilizing one for expertise-based labor markets, since it erodes the premium that expertise commanded by narrowing the gap between the seasoned professional and the well-equipped beginner. Mollick insists on showing both edges of the leveler without softening either.