Talk to It

I spend less time writing code. More time typing in chat windows.

Since I started having AI write code for me, the way I work changed. I used to face the editor and move my hands. Now I write descriptions in a chat field. What I want to build, what state things are in, where I'm stuck. How accurately I convey that determines the quality of the code that comes back.

Imagine a sage sitting by the road. Knows everything, answers anything. But if you can't explain your situation clearly, you get the wrong answer. "Something's not working" doesn't give the sage much to work with.

There was a time when code was communication enough. Commit silently and the message got through. Social awkwardness could be offset by technical skill. That era existed.

What's needed now is the ability to describe a situation from someone else's perspective. Taking the context inside your head and conveying it to someone who doesn't share it. This is a writing problem. Logical, precise, unambiguous prose. It's been given the grand name "prompt engineering," but what you're doing is just composition.

And there's another thing: precise vocabulary. The difference between a select box and a dropdown menu. A modal and a dialog. A button group and a toggle. UI components that look similar but mean different things. Whether you can call them by the right name changes the precision of the code AI returns. Domain knowledge doesn't go away. In the AI era, the power of calling things by their right name matters more, not less.

I don't think the value of writing code has declined. But the value of communicating clearly has risen relative to it. Someone who can code but can't explain, versus someone who can't code but can explain. The latter ships faster in more situations now. Hiring criteria will follow.

I thought I could do both. Then an AI told me it got it wrong because my instructions were ambiguous.