We Stopped Asking AI to Write Code

At first, I used AI the way most people probably do.

I'd describe a problem.

AI would suggest some code.

I'd try it.

Sometimes it worked.

Sometimes it didn't.

It felt productive.

But something was missing.

What changed

During one of the early Wayli engineering sprints, I stopped asking AI to fix problems.

Instead, I started asking it to investigate them.

That changed the conversation completely.

Rather than jumping straight to a solution, we'd spend time understanding the problem.

Where did it first appear?

What evidence supported it?

Was it really a software bug, or was the question itself wrong?

Often the answer wasn't where I expected.

Sometimes the software was working exactly as intended.

The real problem was somewhere else.

AI became much more useful once it stopped acting like a code generator and started behaving like a second pair of eyes.

Why it mattered

That shift changed the way Wayli is built.

Every improvement now begins with understanding the problem before changing the software.

Rather than asking,

"How do we fix this?"

we started asking,

"Why is this happening?"

The goal isn't to produce more code.

It's to understand the system well enough that the right change becomes obvious.

What I learned

Looking back, the biggest improvement wasn't asking AI to write better code.

It was asking better questions.

I've worked as a journalist for most of my working life so it didn't really come as a surprise.

Spark

Inspired by:

• Internal sprint review