The mistakes people make before they start

Before you open any dev tool,
I want to talk about mindset.

Because the traps I fell into
had almost nothing to do with technology.
They were all about mindset.

Mistake one: trying to build every feature at once

This is the most common one, and the most fatal.

"I want to build a platform.
It needs login, search, a rating system,
chat, payments, notifications..."

And then you never finish.

Anthropic (the company behind Claude)
says the same thing in its own engineering write-ups:

Make it work first, then make it work well.

In software this is called an MVP,
Minimum Viable Product.

Which means: build the one feature at the core first,
confirm people want it, then add the rest slowly.

The first version of NaLi Match had three things:
post a request, browse artists, get in touch.
No payments, no ratings, no chat.
But that was enough to connect artists with customers.

Mistake two: adding features that can wait

While you are building, something pops into your head:
"Hey, it would be nice to add feature X here."

So you go build it.
You finish and think of another one, build it, think of another, build that too.

Developers call this scope creep.

Your project keeps growing,
while the core feature never gets done properly.

What works for me: when a new feature comes to mind, write it down on a list. Come back to the list once the core is stable, and you will find more than half of them do not need building at all.

Mistake three: waiting for perfect before you launch

"One more tweak."
"The color of this button is a bit off."
"I am not happy with this layout."

So you keep tweaking, and tweaking,
and three months go by,
and the thing is still sitting on your laptop,
with nobody having seen it.

Mistake four: being afraid of writing bad code

In the AI era this is no longer a problem.

One broken line used to mean a full day of debugging.
Now you paste the error message into AI
and it fixes it in seconds.

Anthropic's own engineering team keeps pointing at the same practice:

Tell AI how to verify its work and let it check its own output. That is one of the most effective things you can do.

So do not be afraid of getting it wrong.
When it breaks, let AI fix it. If it cannot, start that piece over.
Starting over costs almost nothing in the AI era.

Mistake five: building before you talk it through

A lot of people open Claude and say
"build me a website"

and AI does build one,
except it looks nothing like what you pictured,
or you were really just waiting to see what it would pull out of the hat.

Is that still something you made?

The way I think it should go
is to spend time talking it through with AI first:
what you are building, who it is for, what the core feature is,
what tech it uses, how the data gets stored.

That conversation
is what makes your own thinking clear.

Plenty of times you think you know,
and only after talking do you see you never worked it out.

The right mindset

The biggest thing I learned over three years:

Next episode is about talking product architecture with AI,
including frontend, backend and databases, the basics you need to know.