There’s something strange happening with AI development.
For years, software teams had to fight for every feature.
Features were expensive.
They took engineering time.
They needed design.
Testing.
Deployment.
Maintenance.
So people were forced to prioritize.
Now AI can generate features incredibly quickly.
And that sounds like an obvious improvement.
But it creates a new problem:
When building becomes cheap, saying no becomes harder.
“Can we build it?” isn't the useful question anymore
Imagine you're building a booking app.
You start with something simple:
- customer chooses a time
- business confirms the booking
Then you realize AI can add things pretty easily.
So version one becomes:
- user accounts
- staff accounts
- loyalty points
- analytics
- notifications
- subscriptions
- referrals
- reviews
- AI recommendations
- custom dashboards
Technically, maybe you can build all of it.
But you're now answering a completely different question.
You aren't testing:
“Will people use this booking flow?”
You're testing ten assumptions at the same time.
If the product fails, you won't even know which assumption was wrong.
Faster building makes product judgment more important
AI is reducing implementation cost.
It isn't reducing the cost of building the wrong thing.
That cost just appears somewhere else.
You spend time fixing features nobody needed.
Users get a more complicated product.
Your analytics become harder to understand.
The product becomes harder to change.
And suddenly the speed advantage starts disappearing.
This is why I think one of the most valuable features of future AI builders may be something surprisingly simple:
restraint.
An AI builder should probably ask questions before it starts building
Instead of immediately generating 15 screens, I'd rather have the system ask:
Who is this product actually for?
What is the one thing that user needs to accomplish?
What absolutely has to exist on day one?
What can wait?
Who pays?
What assumption are we trying to test?
Those questions aren't as visually impressive as generating a dashboard in 30 seconds.
But they can save days of unnecessary work.
Maybe weeks.
Sometimes the best generated feature is no feature
This is something we're thinking about with built.new.
Full disclosure: I'm involved with the project.
The interesting part to me isn't simply giving someone a prompt box and generating as much software as possible.
It's helping someone move from:
“I have an idea.”
to:
“This is the smallest useful product I should actually build.”
That can mean defining the product before generating it.
Thinking through who it's for.
Planning the important screens.
Separating the things needed now from the things that can wait.
And sometimes pushing back on unnecessary scope.
Because if AI can build ten features quickly, that doesn't mean ten features belong in the MVP.
Small products create cleaner signals
Imagine you launch an MVP with one important workflow.
100 people try it.
60 complete the workflow.
20 come back.
5 pay.
You learned something.
Now imagine launching with 25 different features.
People click everywhere.
Half never finish onboarding.
Some use one feature.
Others use another.
Nobody pays.
What exactly did you learn?
More software does not automatically produce more information.
Often it produces less.
That's why I like this loop:
small build → real user → clear signal → next decision
AI makes that cycle dramatically cheaper.
And that's where I think its real advantage is.
The smartest builder might build less
We've spent a lot of time measuring AI development tools by output.
How much code can they generate?
How many screens?
How quickly?
How many integrations?
Those things matter.
But maybe there's another benchmark:
How much unnecessary product did the tool help you avoid building?
That's harder to show in a demo.
But for someone actually trying to create a business, it might be much more valuable.
The future of AI app building shouldn't only be:
“Tell me what you want and I'll build it.”
It could also be:
“Before we build that — do you actually need it yet?”
And strangely enough, the AI builder that occasionally tells you no might end up helping you ship faster.
