Get the idea out of your head first
AI makes it ridiculously easy to generate software. It does not make it easy to generate clarity.
The tools are addictive. You type one sentence into Lovable, Base44, Bolt, or Replit and suddenly you have dashboards, onboarding, settings panels, pricing pages, and probably a gradient somewhere. It feels like progress immediately. But a lot of the time, the product underneath is still blurry.
Before you define the problem or open a builder, spend time getting the idea out of your head and into a shape you can actually look at.
By the end, you should have:
- a rough sense of who it's for
- the one thing the product is mostly about
- what you're reacting to or trying to fix
- how it should feel
- a short list of things you don't want it to become
1. Remember that ideas are cheap now
Anyone can turn almost anything into a prototype in an afternoon. So the advantage isn't having an idea. It's choosing one worth building.
The trap is that generated UI is now good enough that people confuse screens with clarity. They look at the polished output and think:
The app looks real, therefore the idea must be solid.
Not necessarily.
A useful gut-check before you go further:
Would anyone actually be relieved if this existed? Not impressed. Relieved.
If the honest answer is "I'd be excited to build it, but no one's asking for it," that's a hobby project. That's completely fine. Just don't run it like a business later and wonder why nobody signs up.
2. Know that AI fills in the blanks if you don't
AI builders are optimistic. They'll happily decide things for you.
If you don't define who the product is for, what outcome matters, what the core workflow is, and what kind of experience you're trying to create, the AI starts making those calls on your behalf: navigation, UX patterns, feature depth, workflow structure. Usually the result is something very platform-y. Lots of pages, lots of UI, no center of gravity.
This is why people end up with products they can't fully explain. They didn't really choose the shape of the product. They accepted the default shape the AI generated.
To be fair, that's still useful. If you have no product experience, AI will probably generate something more usable than building every screen from scratch. But there's a real difference between generating software and shaping a product. If your direction is vague, the product usually ends up feeling vague too.
3. Start with a 10-minute voice dump
Before building anything, talk through the idea first with ChatGPT or Claude in voice mode. Literally walk around or drive and ramble for ten minutes.
No structure. No filtering. Just:
- who you think it's for
- the problem you're trying to solve
- things you hate about competitors
- random UX thoughts
- feature ideas
- names
- business model thoughts
- visual references
- edge cases
- things you absolutely don't want the product to become
Voice mode works because it's faster than self-editing. Typing makes people clean up their thoughts too early. Talking gets closer to instinct. The useful stuff is usually buried inside the messy version, and the messy version is faster to produce out loud than at a keyboard.
4. Turn the dump into direction
Once you've done the voice dump, don't immediately open a builder and start generating screens.
First, turn the mess into direction. You don't need a giant PRD. You just need enough structure that the tools understand what matters, what doesn't, what the product actually is, and what kind of experience you're trying to create.
This is where AI is genuinely useful: organizing messy product thinking once there's enough raw material to work with. It struggles when the starting point is "build me an AI app for X." That's when everything turns into platform soup.
A few prompts worth running after a brainstorm:
Find the center of gravity
What is the main thing this product actually does?
If we removed everything else, what workflow still needs to exist for the product to feel valuable?
Spot feature creep early
Looking at this brainstorm, what feels unnecessary?
What feels like roadmap fantasy versus core workflow?
Stress-test the person
Describe the user as a specific person in a specific moment.
What's their week like? Where does the problem hit? What are they doing right before they'd reach for this product?
Once the AI has clearer direction, the outputs improve dramatically. The product starts feeling intentional instead of randomly generated. You're steering now, instead of reacting to whatever the tool invented.
Prompt
This prompt is not here to invent the product for you. It's here to organize the mess into something you can actually build on.
Brainstorm to direction
Take this brain dump and organize it into a clear product direction.
Please structure it into:
1. Core user, as a specific person in a specific moment
2. Core problem
3. The one thing this product mostly does
4. Core workflow
5. First value moment
6. How it should feel
7. Things to intentionally leave out for now
Then audit it.
Please flag:
- where the idea is still vague
- where I'm describing features instead of a workflow
- where this might just be founder curiosity, not a real problem
- what the center of gravity actually is
- what I should sharpen before I keep going
Be direct. If this is still too blurry to build from, tell me.
Before you move on
| Check | What to look for |
|---|---|
| You confused screens with clarity | A polished prototype doesn't mean the idea is solid. Shape it first. |
| You let the tool decide the product | If you can't explain the shape of the product, you accepted the AI's defaults. |
| You skipped the voice dump | The useful stuff is usually buried in the messy, unfiltered version. |
| There's no center of gravity | If everything feels equally important, you haven't found the one thing it's about yet. |
| Only you would be relieved | Excited-to-build is not the same as someone-needs-this. Be honest about which one this is. |