New Release: 'Co-Intelligence' by Ethan Mollick - Mapped

Most books about AI in 2024 had a shelf life of six months. Ethan Mollick's Co-Intelligence is one of the rare ones that's still worth reading a year later, because it isn't really about AI capabilities - it's about how humans should adjust their work patterns to a colleague that doesn't get tired, sometimes hallucinates, and improves quietly every quarter. That framing ages well in any AI moment.

The four rules at the centre

Mollick organises the book around four rules for working with AI. They form the map's primary branches.

Always invite AI to the table. Mollick's argument is that the highest-value use is the use you haven't thought of yet, and the only way to find it is to try AI on tasks you're sure it can't help with. Half the time you'll be right. The other half, you'll discover a use you'd have missed.

Be the human in the loop. AI is a competent intern with an unreliable memory. Treat it that way. Don't outsource judgement; outsource grunt work and reserve your attention for the parts that benefit from a human's accountability.

Treat AI like a person (but tell it what kind of person). Personas matter more than prompt tricks. 'You are a senior editor at The New Yorker who is being asked to review this paragraph' produces dramatically different outputs from 'edit this paragraph.' Mollick spends a chapter on persona engineering and it's the most concretely useful part of the book.

Assume this is the worst AI you'll ever use. The model in your hands is the floor, not the ceiling. Build workflows that compound as the floor rises, rather than workflows optimised for the model's current limitations.

The two big concepts

Around the four rules, two big concepts sit as separate branches:

The Jagged Frontier. AI is unevenly capable. It will solve hard problems and fail at easy ones, in ways that don't map to human intuition about difficulty. The map's job here is to make this concept stickable - it explains 80% of the failure modes you'll encounter in practice.

Four kinds of co-intelligence. Mollick distinguishes AI as person, tutor, coach, and creative partner. Each has different best practices and different risks. This branch is the operational core of the book and should sit close to the four rules.

What the map deliberately omits

Mollick's book contains a lot of cool examples and a lot of speculation about the future. The map should include almost none of it. The examples are useful for the first read but they're not load-bearing. The future speculation will age out faster than the operating principles.

The cleanest Co-Intelligence map has six primary branches (four rules + two big concepts) and roughly fifteen total leaves. If yours has more, you've captured the book's surface but not its spine.

Where the book is weakest

The book is at its weakest when it tries to make grand societal claims. Mollick is a Wharton professor, and his strongest material is on the firm, the team, and the individual knowledge worker. The chapters that try to extrapolate to civilisation-scale change feel rushed. Map this as a small 'where the argument thins out' branch off the side - it's an honest note that will make your re-readings more sceptical and therefore more useful.

How to use the map at work

The four rules are diagnostic. When you find yourself frustrated by an AI tool, run through them: did I invite it (rule 1)? Am I still the human in the loop (rule 2)? Did I give it a persona (rule 3)? Am I building for this model or for the next one (rule 4)? In our experience, 90% of frustration with AI tools is one of these rules being violated.

The map turns the diagnosis into a 15-second lookup. The book turns it into chapters. Both are valuable; the map is what you'll use day-to-day.