Co-Intelligence
- 2024
- First US edition, 2024
- hardcover
- Portfolio (Penguin Random House US)
- 9780593716717
AskNex reference artwork. The publisher's cover appears only under an approved rights record.
by Ethan Mollick
A working guide to using large language models as a collaborator: where they help, where they fail, and how to keep your own judgement in charge.
The most practical account of using AI at work without handing over the parts that make you valuable.
AskNex explains and applies the ideas in this book in its own words. Nothing on this page is quoted from the book, and the page does not stand in for it: the argument, the evidence and the voice are the author's, and reading the original remains the deeper path.
The core idea
Treat AI as a co-worker with a jagged, unfamiliar set of strengths. Invite it into your work, keep a human in the loop, and use it to raise the floor of your output rather than to replace the expertise that lets you judge it.
The ideas
3 mental models, in the order they build on each other.
- 01
The jagged frontier
Capability is uneven: brilliant at one task, hopeless at an adjacent one. You have to map it for your own work.
- 02
Centaur or cyborg
Divide the work between you and the model, or blend it tightly. Either way, never leave the whole task unwatched.
- 03
Always invite AI to the table
Use it on everything once, so you learn where it earns a place and where it does not.
What the book argues
AskNex's own paraphrases of the positions this book takes, as they are held in the knowledge graph. Not quotations, and not the whole book.
Stay the human in the loop: the model is confidently wrong in ways it cannot signal, so checking is part of the work.
Bring the model into every task once, so you learn where it earns a place and where it fails.
Capability is jagged: excellent at one task and poor at an adjacent one, so the frontier has to be mapped for your own work.
Divide the work between yourself and the model or blend it tightly, but never hand over a whole task unwatched.
Working with AI
Using a fast, fallible model as a collaborator whose strengths are uneven and must be mapped.
Human judgement with AI
Keeping the expertise that lets you evaluate what a model produces, instead of outsourcing the judgement.
Career capital
Rare and valuable skills bought with sustained effort, then spent on autonomy.
Cognitive bias
Fast intuitive judgement is efficient and predictably wrong in known ways.
AI-assisted software engineering
Specifying, steering and verifying code produced by a model, rather than typing it.
Where it argues with other sources
Most summaries flatten this. AskNex keeps it, because the disagreement usually decides what applies to you.
Mollick qualified by Kim & Yegge
Expertise is what lets you judge the output; skipping the apprenticeship to rely on the model leaves you unable to evaluate it.
Refusing these tools out of pride or fear now costs more than learning them badly would.
Adopt the tools now, but not in place of the apprenticeship: use them where you can already judge the output, and build the fundamentals unaided where you cannot.
Newport qualified by Mollick
Concentration is becoming scarce, which is precisely what makes it valuable.
Bring the model into every task once, so you learn where it earns a place and where it fails.
Inviting the model into everything is how you map it; concentration is what the model cannot supply. Run the experiment, then protect the deep blocks from it.
What actually matters
- Expertise is what lets you evaluate the output. Skip the apprenticeship and you lose the ability to judge.
- The model is confidently wrong in ways it cannot signal; verification is part of the job.
- Context and persona shape the result more than clever wording.
Where it applies
Practical moves the book supports. Each one is a place to start, not a rule.
- Map the frontier
For one week, run every task past the model once and record where it helped, hurt or hallucinated.
- Keep the check
Decide in advance which outputs you will verify line by line and which you will only skim.
- Protect the apprenticeship
Name the skills you still need to build unaided, and do those without it.
Questions worth asking
Which parts of my work would I be unable to evaluate if the model did them?
Where has AI quietly become the author of decisions I still sign?
Problems this book speaks to
Recurring situations where this book is one of the sources AskNex draws on. Each opens a public orientation page.
What are you trying to figure out right now?
AskNex will answer your situation using the ideas in Co-Intelligence alongside every other source that applies — including the ones that argue against it.