AI is not taking my job (yet)

I've been writing software for a long time, and yes, I've had the thought: is AI going to take my job?

So I tried to find out. I picked a small but real project and set out to vibecode it end to end with Claude Code. No hand-holding. Describe the task, let it build, see what comes out.

Some of it genuinely impressed me. It made a few clever decisions I hadn't put in the task description at all. Sensible defaults, a couple of edge cases handled before I noticed them.

And then I opened the code.

CSS as inline styles, everywhere. Classes with no clear module structure, just things that happened to end up next to each other. And the business logic had flaws, the kind that don't show up in a demo but would show up in a month, with a customer's money involved.

It worked, sort of. It wasn't something I'd want to maintain, and I definitely wouldn't want to be the person inheriting it.

What struck me is how different this was from my day-to-day. Inside an existing project, with established structure, conventions, guidelines and a decent harness around the model, the same tool is excellent. It follows the patterns that are already there, it stays in its lane, and it saves me hours every week.

Give it a blank page and a paragraph of intent, and there are just too many decisions with nothing to anchor them. Every one of those is a place where things can go quietly wrong.

So my conclusion, for now:

I'm not worried about my job. I'm a bit worried about the projects that get built this way and end up on someone's desk two years from now.

If one of those has already landed on your desk, that's a conversation we have often.

All posts