Author: Ethan Mollick · Published: 2024 · Genre: AI, Business
Co-Intelligence‘s biggest strength is its practical, evidence-informed advice from a Wharton professor who experiments with AI constantly and helped run one of the best-known studies of AI at work. Its biggest weakness is that it is a snapshot of a fast-moving technology: many examples dated within months, and it moves quickly past harder questions about risk, jobs and who benefits. It’s worth reading for professionals, teachers and managers who haven’t yet made AI part of their daily work.
The core ideas
- Four rules. Always invite AI to the table, meaning try it on everything to learn what it can do. Be the human in the loop, checking its work and judgment. Treat AI like a person, but tell it what kind of person to be. And assume this is the worst AI you will ever use.
- The jagged frontier. AI is excellent at some tasks and poor at others that look just as hard, and the boundary is invisible until you test it. In a study Mollick co-authored with Boston Consulting Group consultants, those using GPT-4 finished about 12 percent more tasks, about 25 percent faster and at higher quality on tasks inside the frontier. On a task designed to fall outside it, AI users were less likely to reach the right answer than those working alone.
- Centaurs and cyborgs. Centaurs divide the work, handing some tasks to AI and keeping others. Cyborgs weave AI into their work continuously.
- Roles for AI. Chapters cover AI as a person, a creative partner, a coworker, a tutor and a coach, with concrete ways to use it in each role.
- Four futures. Mollick sketches scenarios ranging from AI staying roughly as good as it is now, through slow or exponential growth, to a machine god. He argues the middle scenarios of continued improvement are the ones to plan for.
What’s great
The advice is specific and usable. Give the AI context and a role, ask for many options rather than one, push back on its answers and keep iterating. Readers who have only asked a chatbot one-off questions will come away working differently.
The jagged frontier is the book’s most valuable idea. It explains why AI can write a sonnet with ease yet struggle with a poem of exactly fifty words, and why the only way to find the boundary is hands-on experimentation. The same study found that lower-performing consultants gained the most from AI, a striking result for anyone thinking about training and skills.
Mollick is also honest about limits. He explains that AI makes things up confidently, so its output always needs checking, and he warns that using AI for first drafts could erode the apprenticeship through which junior people learn.
What’s weak
It dates quickly. The book was written around the GPT-4 era, and many of its examples already feel old next to newer reasoning models and AI agents. Mollick’s fourth rule anticipates this, but it still limits the book’s shelf life.
The advice to treat AI like a person is practical, yet contested. Critics argue that anthropomorphizing chatbots encourages over-trust and emotional attachment. Mollick flags the risk but leans into the framing anyway.
The book is also light on the big picture. Reviewers concerned about AI safety note that it sets aside catastrophic risk quickly, and it says little about job losses or who captures the gains. Much of the evidence comes from early studies and working papers, including Mollick’s own, whose results may not hold across other jobs and newer tools.
Who should read it
Read it if you are a professional, educator or manager who wants a fast, sensible start with AI. Experienced users will find less that’s new, and should follow Mollick’s newsletter, One Useful Thing, for more current advice.
Rating: 4/5
Image credits — Featured image: cover image via Open Library; cover design © the publisher, shown for review purposes. In-article images: Photo by Tara Winstead on Pexels.
