A friend complained to me recently: the product requirements analysis he spent two weeks on, AI generated a more comprehensive version in three minutes. He was a bit panicked: “Where’s my value?”

I said: If your job is “organizing requirements into documents,” then yes, that’s risky. But what if you can discover needs that even the customer hasn’t realized yet?

In the AI era, the division of labor between humans and machines is being redrawn. What we should train most isn’t “solving problems faster than AI,” but three things AI can’t do yet.

First Muscle: From “Solving Problems” to “Discovering Problems”

What AI Excels At: Optimal Solutions to Given Problems

Give ChatGPT a clear question: “How to optimize this code’s performance?” It’ll give you 10 optimization approaches with benchmarks.

Give Claude a specific task: “Write a market research report analyzing electric vehicle industry trends.” It’ll generate a professional analysis with latest data and clear logic.

AI’s prerequisite: the problem is already clearly defined.

Human Advantage: Discovering Hidden Real Problems

In 2007, Steve Jobs didn’t ask “how to make phone keyboards better,” but “why do we need keyboards?” Then iPhone was born.

Tesla wasn’t optimizing “how to make cars more fuel-efficient,” but questioning: “Why must cars burn fuel?”

Airbnb founders weren’t thinking “how to make better hotels,” but saw “every home has unused rooms.”

These questions, AI can’t generate. Because they don’t come from “summarizing existing data,” but from dissatisfaction with the status quo, deep user understanding, and keen trend sensing.

How to Train This Muscle?

1. Ask “Why” Instead of “How”

  • ❌ Low-level: “How to improve user retention?” (execution question, AI can answer)
  • ✅ High-level: “Why do users churn? What do they really want?” (insight question, requires deep thinking)

2. Observe “Pain” and “Inefficiency”

All great products stem from pain points. Observe:

  • What are people complaining about?
  • Which processes are tedious but everyone accepts them?
  • Which needs are ignored by existing solutions?

AI can analyze data, but it can’t “feel pain.” You can.

3. Cross-Domain Connections

Uber is “taxi + internet” combined.
Netflix is “movies + big data recommendations” combined.
Notion is “notes + database” combined.

AI is strong in single domains, but cross-domain metaphors and analogies require human life experience and intuition.

Engage with different fields, think about “can solution from field A apply to field B?”

Second Muscle: From “Executing Tasks” to “Creating New Species”

What AI Excels At: Optimizing Existing Paths

AI can write code, but it writes “already-validated code patterns.”
AI can draw, but it draws “styles existing in training data.”
AI can write articles, but it writes “expressions from the corpus.”

AI is the strongest “efficiency engine,” but not a “creation engine.”

Human Advantage: Creating Unprecedented Things

Picasso created Cubism—before that, no training data could generate this style.
Radiohead created the electronic rock of “Kid A”—before that, no similar music existed.
Elon Musk decided to build reusable rockets—before that, everyone thought it was a crazy idea.

Creativity isn’t “optimizing the existing,” but “zero to one.”

How to Train This Muscle?

1. Boldly Try “Unreasonable” Ideas

AI avoids risk because its goal is “high-certainty output.” But breakthroughs often come from “unreasonable experiments.”

  • PayPal’s original idea was “payments between PDAs” (sounded dumb)
  • Twitter’s original idea was “140-character status updates” (people thought no one would use it)

Allow yourself a 90% failure rate—that 10% might change the world.

2. Mix and Recombine

Innovation isn’t magic, but combining existing elements in new ways.

  • iPhone = phone + iPod + internet browser
  • Slack = chat tool + workflow integration + search engine
  • Notion = documents + database + collaboration tool

Ask yourself: Can I combine A and B in a way no one’s tried?

3. Don’t Fear “No Answer” Exploration

AI needs clear objective functions, but many creative works start with “not knowing what I’ll create.”

Van Gogh didn’t think “I’ll paint an $80 million painting” when creating “Starry Night.”
The Beatles didn’t think “I’ll make an album that changes rock history” when recording “Sgt. Pepper’s.”

Reserve time for “purposeless exploration”—don’t let “task-driven” fill all your time.

Third Muscle: From “Improving Efficiency” to “Raising Capability Ceiling”

What AI Excels At: Amplifying Your Ability 10x

You can code, AI makes you code faster.
You can design, AI makes you iterate faster.
You can write, AI makes you produce more.

AI is a “multiplier,” but it multiplies your existing capabilities.

If your capability ceiling is 10, AI helps you reach 100.
If your capability ceiling is 1, AI helps you reach 10.

The key is: Where is your capability ceiling?

Human Advantage: Breaking Through the Ceiling

In 2016, AlphaGo defeated Lee Sedol. Interestingly, after studying AlphaGo’s games, human Go players discovered many moves they’d never considered before.

Top players’ skills actually improved because of AI, because they learned to see Go from new perspectives.

AI can be a coach, but the drive to breakthrough comes from humans.

How to Train This Muscle?

1. Actively Leave Your Comfort Zone

Efficiency tools make you more efficient in your comfort zone, but capability breakthroughs require stepping out.

  • Learn a completely unfamiliar skill (not for practicality, but to break mental patterns)
  • Challenge a project you think “I can’t possibly do”
  • Deep engage with people 10x better than you

AI can help you do things faster, but can’t help you choose “what to do.”

2. Embrace Failure and Confusion

AI training aims to “reduce loss function,” while human growth is essentially “learning from failure.”

If you only do things AI can help you complete perfectly, you’ll never breakthrough.

Allow yourself imperfect attempts, allow yourself 6 months of “producing nothing” exploration.

3. Cultivate “Meta-Cognitive Ability”

AI can give you answers, but you need to know:

  • Is this answer right?
  • Is this direction worth pursuing?
  • Are there blind spots in my thinking?

Meta-cognition is “thinking about your thinking”—this is where AI is currently weakest.

Ask yourself often:

  • Why do I think this way?
  • What are my assumptions?
  • Am I trapped in some mental pattern?

Practice: A Comparison Case

Suppose you want to create a fitness app.

Low Capability Ceiling Approach (AI Can Replace):

  1. Check existing fitness apps
  2. Analyze feature lists
  3. Find designers and developers to outsource
  4. Use AI to generate copy and materials
  5. Launch a product “similar to others”

→ This is “executing tasks”—AI + low-code tools can do it.

High Capability Ceiling Approach (AI Cannot Replace):

  1. Discover hidden problems: Spend 3 months at gyms, deeply talk with 50 fitness newbies, discover “people don’t lack knowledge about how to train, but can’t persist”
  2. Create new species: Don’t make a “fitness instruction app,” but a “fitness social gamification app,” turning fitness into an experience with instant feedback and achievement like gaming
  3. Raise capability ceiling: Learn game design, behavioral psychology, community operations, use cross-domain knowledge to redefine fitness apps

→ This is “creating new possibilities”—only humans can do it.

Conclusion: In the AI Era, Be a “Question Asker” Not an “Answer Giver”

For the past 200 years, the Industrial Revolution trained us to be “efficient answer givers”:

  • Boss gives you problems, you give answers
  • Exams give you questions, you give solutions
  • Market gives you demands, you give products

In the AI era, the value of “answer givers” is rapidly declining.

What’s truly scarce are those who can ask good questions, create new species, and break through capability ceilings.

Don’t compete with AI on speed, compete on direction.

Machines can run faster, but only humans can decide where to run.

Reflection: In your recent work, how much is “irreplaceable by AI”? If the proportion is low, maybe it’s time to rethink your positioning.