Co-Intelligence Summary: How to Work With AI, Not Against It

Picture a Monday report with one wrong number in it. You did not write that number. An AI tool did, and it sounded completely sure. Most people ask whether AI will take their job. Ethan Mollick asks a more useful question in Co-Intelligence: which parts of my work should stay mine?

What is Co-Intelligence about?

Ethan Mollick is a professor at the Wharton business school who studies how people work with AI. His book, published in 2024, is not a technical guide. It is about what happens when ordinary professionals start using tools that write, analyse and reason.

His central idea is simple. AI is not a human and it is not normal software. Treat it as a coworker who is very talented and often wrong. That coworker can lift your work a great deal, but only if you know where it can be trusted.

Why AI takes tasks, not jobs

Many people picture AI taking a whole job at once. Mollick describes something smaller. A job is a bundle of tasks, and AI takes some tasks and leaves others.

The book cites research that compared what AI can do against more than a thousand kinds of work. Only 36 of 1,016 job types had no overlap with AI. These were mostly jobs done with the body, like roofer or dancer. The most exposed jobs were creative and highly educated ones.

Think of a senior analyst in Bengaluru who writes a market report every week. Her job is not “report writer”. It is data cleaning, first drafts, formatting, thinking, and talking to the client. AI may take the first three. The last two are still hers.

The jagged frontier: where AI stops being good

This is the most useful idea in the book. Mollick asks you to picture a castle wall seen from above. Some towers stick far out, and others fold inward. The wall is what AI can do well. He calls it the jagged frontier.

Inside the wall, the work is excellent. Outside it, the work falls apart. You cannot see the wall, and tasks that feel equally hard can sit on opposite sides. AI can write a lovely sonnet but may fail at a poem of exactly fifty words.

The book describes a study of nearly 800 consultants at Boston Consulting Group. Those with AI were faster and produced better work on most tasks. But on one task, built to sit just outside the frontier, they did worse. Without AI, 84 percent got it right. With AI, only 60 to 70 percent did. The AI gave an answer that sounded right, and good people trusted it.

The lesson is plain. Hard for you does not mean hard for AI. You find the edge only by testing.

Why AI acts like a person, not a search engine

Under the surface, AI does not look facts up. It learned from huge amounts of human writing, and it predicts which words fit next. That is why it is so good at pitches, emails and polite refusals. It is also why it can invent things that sound true.

Mollick mentions a lawyer who filed a court brief with six cases that did not exist. A chatbot had made them up. Such mistakes are hard to catch because they read so smoothly.

The book also shows that AI answers change with how you treat it. Ask it like an opponent and it argues. Ask it like a teacher and it explains. This is why Mollick says to treat it like a person, even though it is not one.

Why the better the AI, the more you must stay alert

Many of us assume a stronger tool means fewer mistakes to worry about. The book shows the opposite risk. In a study of recruiters, those given a better AI did worse than those given a weaker one. They spent less time on each application and followed the machine blindly. Mollick calls this falling asleep at the wheel.

The book also finds that AI helps weaker performers the most. A call centre study showed the lowest performers becoming 35 percent more productive, while experienced staff barely changed. If your edge was being faster than the person next to you, that edge has just become cheaper.

Four rules for working with AI

Mollick gives four rules.

1. Always invite AI to the table. Try it on everything you do, unless law or ethics stop you. Nobody can map the edge in your job except you.
2. Be the human in the loop. Check the work and own the result. You are never just the person who presses send.
3. Treat AI like a person, but tell it what kind of person. A few words typed like a search box give you an average answer. Give it a role and context.
4. Assume this is the worst AI you will ever use. The tools improve fast, so a task that failed last year may work now.

How to use this at work and at home

1. Write down every task from your last week. Pick one you have never tried with AI and try it today.
2. Give the tool a role in each request. For example, ask it to act as a doubtful client and list the three claims in your report it would challenge first.
3. Check every figure, name and date against its source before you share it. Do this yourself, every time.
4. Sort your tasks into three piles. Just Me is work you keep. Delegated is work AI does and you check. Automated is work AI does alone. Mollick says the third pile is small for now.
5. Review the sort once a month, because some tasks will have crossed the frontier.

Our take

The strongest part of this book is the jagged frontier. It gives you a practical reason to experiment instead of either fearing AI or trusting it blindly. The consulting and recruiter studies are also a good reminder that mistakes grow when we stop paying attention.

The weaker part is that AI changes quickly. Some examples will feel dated soon, and the book cannot tell you where the frontier sits for your job today. It is also written mainly for office workers in knowledge jobs. If your work is mostly physical or deeply local, the advice applies less directly.

One warning in the book deserves more attention than it gets. If AI does the basic work that juniors once did, how will they ever learn to spot its errors? If you manage people, keep some of that work with them and let AI act as a coach, not a ghostwriter.

The animated summary walks through each idea with a story about one analyst and her three piles. It takes about 20 minutes.