Co-Intelligence
Ethan Mollick · 2024
Editorial rating
- Evidence
- 9/10
- Actionability
- 8/10
- Originality
- 7/10
The thesis
AI isn't coming - it's here. The real question isn't whether AI will change work, education, and creativity, but how quickly you'll learn to work with it rather than fear it. We've built the first generally applicable intelligence-augmentation technology, and those who treat AI as a co-worker, co-teacher, and creative partner will thrive.
Who this is for
Knowledge workers, educators, managers, and creatives who need practical guidance on integrating generative AI into daily work. Anyone paralyzed by AI hype or fear who wants evidence-based clarity on what works today.
My favorite quote
You should try inviting AI to help you in everything you do, barring legal or ethical barriers. Familiarizing yourself with AI's capabilities allows you to better understand how it can assist you - or threaten you and your job.
Why it matters
Most people either ignore AI entirely or use it once and dismiss it. Mollick argues systematic experimentation is the only way to understand the "Jagged Frontier" of what AI does well vs. poorly.
Do this
Pick one task you do weekly. Spend 30 minutes today asking ChatGPT or Claude to help you with it. Document what works, what fails, what surprises you.
Start here
AI hallucinates, has no consciousness, and makes mistakes - but used correctly, it's already more powerful than most people realize. The "Jagged Frontier" metaphor is key: AI excels unpredictably at some tasks (writing marketing copy, brainstorming, coding boilerplate) and fails catastrophically at others (legal citations, precise math, nuanced judgment). Your job is to map that frontier through experimentation, then delegate ruthlessly to AI what it does well while keeping human judgment where it matters.
Critical summary
Mollick, a Wharton professor who teaches the first AI-required MBA course, became a leading explainer of generative AI through his One Useful Thing newsletter. This book distills his early hands-on experience with GPT-3, GPT-4, and other LLMs into practical frameworks for work, education, and creativity.
Core structure: AI as Co-Worker (automating/augmenting knowledge work), AI as Co-Teacher (personalizing education at scale), AI as Creative Partner (ideation, iteration, execution). Mollick emphasizes the "Homework Apocalypse" in education and provides tactical solutions: requiring AI use with fact-checking, designing AI-resistant assessments, using AI as tutor/coach.
What it gets right
- Evidence-based: Cites studies showing professionals using AI complete tasks 25-40% faster with higher quality ratings
- Practical frameworks: The "Jagged Frontier," Level 1 vs. Level 2 decisions (reversible vs. irreversible), treating AI as "weird alien intelligence"
- Education-focused: Two full chapters on AI in teaching/learning with real examples from his Wharton courses
- Honest about limitations: Hallucinations, lack of explainability, bias, job displacement risks all discussed
- "Use it everywhere" philosophy: Only way to discover where it helps vs. hinders
What it misses
- Recency bias: Mollick's examples skew toward companies/students struggling with AI adoption, not organizations doing it well
- Light on enterprise implementation: More startup/individual focus, less on change management at scale
- Optimism bias: Downplays structural barriers (poverty, access, digital literacy) for AI adoption
- Shallow on AI risks: Brief mentions of bias/misuse, but doesn't grapple with existential risks or regulatory challenges
- Fluff complaints: Goodreads reviewers note repetitive sections and excessive praise for basic concepts
Evidence quality: Strong. Multiple peer-reviewed studies cited (Harvard/BCG consultant study, Wharton MBA experiments). But most evidence comes from 2022-2023 (GPT-3.5/GPT-4 launch era), which is already outdated given GPT-4o, Claude 3.5, etc.
Critical reception: NYT Bestseller. Goodreads 4.09/5 (4,000+ ratings). Praised by Reid Hoffman, Dan Pink, Angela Duckworth. Critics say it's "AI cheerleading" without sufficient skepticism. Some call it "extended blog post" (200 pages, large print, could be 100 pages). Strong for beginners, obvious for practitioners already using AI daily.
Key concepts
The Jagged Frontier
AI's capabilities are unpredictable - it crushes some tasks and fails others without logic. Map it through experimentation before trusting AI with critical work. Test weekly: what AI tasks became possible this month that weren't last month?
Treat AI Like a Person (But Remember It's Not)
Best interaction style is conversational - give context, assign roles, iterate. But it has no consciousness, emotions, or understanding. It's predicting text, not thinking. Use the "pompous helpful assistant" or "critical reviewer" prompts to get different perspectives.
The Homework Apocalypse
Traditional essays are dead - AI writes them better than most students. Solution: Require AI use, focus on fact-checking, metacognition, and application rather than generation. Redesign assessments around AI-resistant skills.
Level 1 vs. Level 2 Decisions
(from Bezos): Level 1 = hard to reverse (hiring, firing, major investments). Be slow, cautious. Level 2 = easy to reverse (testing content, iterating). Move fast, use AI, don't wait for perfect data. Most work decisions are Level 2.
Co-Intelligence, Not Replacement
AI augments human judgment, doesn't replace it. Use AI to generate first drafts, brainstorm options, challenge assumptions - then apply human expertise to evaluate, refine, decide. Cyborg > AI alone > Human alone (for many tasks).
Hallucinations Are Fundamental
AI doesn't "know" anything - it generates probabilistic text. Always fact-check outputs, especially citations, numbers, names. Hallucinations won't be "solved," they're intrinsic to how LLMs work.
Core insights
-
"We have created a weird alien mind - one that isn't sentient but can fake it remarkably well"
Stop anthropomorphizing. It's not thinking, it's autocomplete on steroids. This matters for trust calibration.
-
AI amplifies whoever uses it
High performers get more leverage, low performers improve more (studies show bigger gains for weaker workers). But skill gaps persist - AI doesn't make you a genius, it makes you a faster version of yourself.
-
The "Jagged Frontier" changes weekly
What AI couldn't do last month it might do this month. What it does well today might regress tomorrow. Continuous experimentation is mandatory.
-
Personal tutoring gives students a two-sigma advantage (98th percentile)
AI tutors can scale this for free. Education will be revolutionized - question is whether we embrace it or fight it.
-
Boredom is dangerous - AI alleviates it
Studies: 66% of men, 25% of women shock themselves rather than sit bored for 15 minutes. AI as always-on creative partner reduces harmful boredom behaviors (doom-scrolling, etc.). But also risks over-reliance.
-
"You can't figure out why an AI is hallucinating by asking it"
The AI will generate a plausible explanation, but it's bullshit. LLMs have no introspection - they're pattern-matching machines.
-
Prompt engineering is overrated
More important: domain expertise, critical thinking, ability to evaluate AI output. Teach students to be "humans in the loop," not prompt wizards.
Implementation steps
Today
- Pick one repetitive task (email, report, code, analysis). Ask AI to do it. Compare quality to your usual work.
- Create an "AI Experimentation Log" - track what you try, what works, what fails. Goal: 10 experiments this month.
This week
- Use AI for 3 different use cases: (1) Creative (brainstorm, write), (2) Analytical (research, summarize), (3) Executional (code, format, polish)
- Identify your "Jagged Frontier" - which tasks should you delegate to AI vs. keep human?
- Test AI as a coach: Ask it to challenge your assumptions on a decision you're making
This month
- Integrate AI into daily workflow: Set up Custom Instructions in ChatGPT or Projects in Claude with your role, goals, preferences
- If you manage a team: Run a workshop on AI experimentation. Share successes/failures. Normalize learning through failure.
- Redesign one recurring task to be "AI-first" - let AI do first draft, you refine
Ongoing
- Weekly review: What did AI help with this week? What failed? Where is the frontier shifting?
- Stay current: GPT-5, Claude 4, Gemini Ultra coming soon. Revisit your "AI can't do this" list quarterly.
- Teach someone: Best way to solidify your AI knowledge is to coach a colleague
Suggested 30-day practice plan
An editorial application plan created by Monolithic Vault - an interpretation of the book's ideas, not part of the original book.
- Day 1
Read Co-Intelligence (or key chapters: AI as Co-Worker, AI as Creative). Get the frameworks.
- Day 3
Set up your AI tools - ChatGPT Plus, Claude Pro, or free versions. Create your first custom prompt.
- Day 5
Use AI for a work task you've been avoiding. Did it help? Document it.
- Day 7
Try AI as tutor - ask it to explain a concept you're weak on. Compare to Googling.
- Day 10
Use AI for creative work - brainstorm 20 ideas for a project. Pick the best 3. Would you have thought of them without AI?
- Day 14
Mid-point reflection - what surprised you? What disappointed you? Recalibrate your frontier map.
- Day 21
Teach someone how you're using AI. Articulating your process clarifies it.
- Day 30
Write a "State of My AI" memo - what's now delegated to AI, what you'll never delegate, what you're still experimenting on.
Free PDF summary
Take this analysis with you: a designed two-page field-notes sheet with the thesis, my favorite quote, the key concepts and core insights, and the full 30-day checklist. Print it or keep it - free, no signup.
Go deeper
If this analysis earned your attention, the full book goes further than any summary can. The original is always the primary source.
