1. You hired 100 geniuses. They're all in the break room because nobody gave them a task
Generative AI is wild. It writes code. It writes copy. It does research. It translates. It has read more than any of us ever will.
Hear that list and you'd think everyone would be running 10 side businesses by tomorrow.
In real life, most people stall right here:
"Okay... so what do I actually build?"
That one question.
The smartest AI in the world does nothing if the to-do list is empty. Line up 100 Ferraris with no destination and all you've got is a very fancy parking lot.
What's missing in the AI era often isn't a better model.
It's someone to hand that power its first job.
2. You don't need a world-changing invention. You need the first spark
When people hear "good at ideas," they picture someone who suddenly dreams up a revolutionary invention nobody has ever thought of.
The skill that actually matters is plainer, and stronger:
- "This part is a pain."
- "Could these two things work together?"
- "Could this feature be used for something else?"
- "Wait, why do we even do this step?"
- "What would we learn if we just built the smallest possible version?"
That's the skill: creating sparks.
Your first idea doesn't have to be finished. What matters is giving the next thought something to stand on.
Nobody can react to an empty table. Put even a rough prototype on it, and suddenly everyone has something to say: "This is weird." "This part is useful." "Then let's try this next."
The first move isn't the answer.
It's a machine that churns out the next answers.
3. "Being creative" is actually a bunch of different skills
Break it down and you get at least these parts:
| Skill | What it looks like |
|---|---|
| Divergence | Coming up with several options from one problem |
| Analogy | Borrowing how something works in a different field |
| Recombination | Taking existing parts A + B and making C |
| Problem finding | Spotting the annoyance that comes before the problem you were handed |
| Reframing | Asking "what are we really trying to solve?" |
| Convergence | Picking which idea to actually build |
| Breaking it down | Splitting the work into chunks AI can handle |
| Starting | Just making the first real thing |
| Re-thinking | Looking at what works and getting the next idea |
What psychologists call divergent thinking is an important piece of this. The APA describes divergent thinking as creative thinking that solves problems or makes decisions using strategies that break from the usual or previously taught ones.[1]
But there's a catch: scores on divergent-thinking tests don't capture real-world creativity as a whole.[2]
So being able to list a lot of ideas isn't enough.
The real game is reframing, choosing, building, and thinking again.
4. Creativity is a back-and-forth: generate, explore, generate again
Research on creative thinking has a model for this: you don't produce a finished idea in one shot. You generate rough candidates, explore and judge them, and go back to generating when needed. It's called the Geneplore model.[3]
Map that onto building with AI and it clicks right away:
- Get an idea
- Bounce it off the AI
- Turn it into a rough spec
- Build it
- Run it
- Realize "hmm, not quite"
- Find a different angle
- Build it again
In other words, your idea doesn't need to be complete before you start building.
Building is part of the thinking.
That's the big change.
When a prototype takes 3 months, you want to think everything through in your head first.
When a prototype takes 30 minutes, the prototype itself becomes your notes.
"Ideas people" and "builders" stop being separate departments.
5. "I can't multitask" is true, if you're doing everything yourself
"I can't juggle that many things at once."
Fair enough.
Try holding the specs, code, research, checks, and fixes for 10 projects in your head at the same time, and every tab in your mental browser goes blank.
So change the setup instead.
You don't have to multitask. Let the system run things in parallel.
Your job is just:
- What do I want?
- What matters most?
- What's acceptable and what isn't?
- Which result do I go with?
After that, split up
- research
- writing the spec
- building
- testing
- translating
- reviewing
into pieces that can run on their own, and hand them to AI.
"I'm bad at multitasking, so AI isn't for me" gets it backwards.
If you're bad at multitasking, getting the work out of your head and turning it into clear, separate jobs is worth even more.
6. All that AI knowledge is a warehouse of materials, not a blueprint
No matter how much the AI knows, that alone doesn't tell you what to build.
Picture a giant hardware store with every kind of lumber, steel, glass, tool, and heavy machine you could want.
Amazing.
But until a customer decides what to build, it's still just a hardware store.
AI is similar.
It can pull up information. It can compare options. It can write code. It can pitch ideas.
But the moment someone says:
"I want this annoyance gone." "It'd be fun to connect these two." "I want to take this experience up a notch."
that sense of direction turns raw ability into actual work.
You don't need to know more than the AI to get value out of it.
You need to be able to tell it where to dig.
7. Before solving the problem, the winners rethink the problem
Design research has long treated "framing" as a key practice for messy, open-ended problems. Kees Dorst focuses on reframing: changing how an organization looks at a tough problem situation in the first place.[4]
That matters even more with AI.
Ask AI to "do this," and it may solve exactly what you asked, incredibly fast.
That's the scary part.
If the problem is wrong, you can now mass-produce convincing wrong answers at top speed.
The really useful questions don't stop at:
"How do we make this task faster?"
They go back further:
"Do we even need this task?" "Does the user want this task faster, or gone completely?" "Could we solve this from a totally different angle?"
The better AI gets at answering, the more leverage you get from asking the right question.
8. Building faster gives you more ideas, and that's a quietly huge deal
There's experimental evidence that generative AI can save time and improve quality on certain kinds of knowledge work. In a study by Noy and Zhang, people doing professional writing tasks took 40% less time on average, and their work was rated 18% higher in quality.[5]
Of course, that doesn't mean every job gets 40% faster.
The bigger point is something else:
When trying things gets cheaper, you can try more things.
And the more you try, the more you learn from the real thing:
- Is it actually useful?
- Does it look worse than I imagined?
- Would it work better for something else?
- Could it connect to the feature next to it?
- Where do users get confused?
That kind of info is hard to come up with just by thinking.
So the shorter the loop of
idea → build → watch → new idea
the more your ideas turn into a chain reaction instead of one-offs.
Building speed isn't just productivity.
It's how fast your idea engine spins.
9. But if AI comes up with all your ideas, everyone ends up climbing the same mountain
AI can help with brainstorming.
But in an experiment by Doshi and Hauser, short stories written with AI-generated ideas were rated higher on average, while the AI-assisted stories became more similar to each other, so the group as a whole became less diverse.[6]
That's important.
If you just ask AI for "10 fun ideas" every time, you might end up on the same tour bus as everyone else.
That's why your human input still matters:
- things that genuinely annoyed you
- weird stuff you noticed on the job
- the quirks of your existing system
- comparisons borrowed from another field
- odd combinations
- the pushback: "Sure, that's how everyone does it, but is it right?"
This kind of input only you have gives the AI's giant pool of knowledge a direction and a point of view.
Don't ask AI to do all the thinking.
Toss in one strange pebble yourself, and let AI amplify the ripples.
It's more fun that way.
10. The best loop: notice → one sentence → split → hand off → check → notice again
In practice, it can be this simple:
① Spark "Wouldn't it be handy if this worked like that?"
② One sentence Write down what you want to change, for whom, and how, in one sentence.
③ Split it up Break it into separate jobs: research, design, building, checking.
④ Hand it off Give those jobs to AI. Don't hold everything in your head at once.
⑤ Make something real Text, a screen, code, results: anything you can actually see.
⑥ Read it back Read it, click through it, test it. Did you really get what you were aiming for?
⑦ Next idea "If this works, then we could also do that."
Step ⑦ is why it never ends.
Each thing you build becomes material for the next idea.
You build feature A. Because A exists, you think of B. Once B exists, A + B = C. Once C exists, you start wondering, "Could we redefine the whole service as D?"
It started as a tiny spark, and suddenly there's a power plant.
11. The trap: ideas coming faster than you can organize them
This approach has an obvious side effect.
When ideas keep coming and building is fast, your system grows add-on after add-on.
Then one day you open up the wiring.
Who built this mess?
You did. Yesterday.
The fix isn't to stop having ideas.
- Don't create too many "official" sources of truth
- Don't build two systems that do the same job
- If it can be merged into something that already exists, merge it
- Only run truly independent tasks in parallel
- Always check the result after building it
- Put quality checks in front of important changes
- Separate "we could build this" from "we should build this now"
Fast ideas need fast cleanup.
Don't ease off the gas. Get better steering and brakes.
12. Bottom line: the edge isn't "using AI," it's "giving AI work to do"
AI models will keep getting better.
They'll know more. Code and writing will get faster. Running several AIs at once will get easier.
So the rare human skill slowly shifts.
From "I can build everything myself"
to:
Finding what's worth building.
Lighting the first spark.
Rethinking the problem.
Splitting up the work.
Handing it to AI.
Looking at the real result.
Getting the next idea from there.
AI is a huge amount of ability.
But that ability just sits there until someone gives it a job.
So the people who win aren't only the ones who code faster than AI.
They're the ones who can say, "Wait, couldn't we build this?"
In a company that just hired 100 genius AIs, the first person you need isn't genius number 101.
It's the person who writes the first line on the to-do list.
References (6)
- APA Dictionary of Psychology, “divergent thinking. dictionary.apa.org
- American Psychological Association, “The science behind creativity” — notes that divergent-thinking measures are useful but do not map cleanly onto real-world creativity as a whole apa.org
- Ward, T. B., Smith, S. M., & Finke, R. A., “Creative Cognition,” in Handbook of Creativity; see also Geneplore descriptions of iterative generative and exploratory processes cambridge.org
- Dorst, K. (2011). “The core of ‘design thinking’ and its application.” Design Studies, 32(6), 521–532 doi.org
- Noy, S., & Zhang, W. (2023). “Experimental evidence on the productivity effects of generative artificial intelligence.” Science, 381(6654), 187–192 doi.org
- Doshi, A. R., & Hauser, O. P. (2024). “Generative AI enhances individual creativity but reduces the collective diversity of novel content.” Science Advances, 10(28), eadn5290 doi.org
