Vibe coding solves the wrong problem

Vibe coding is everywhere right now. It will replace software development as we know it, supposedly. Writing code is superfluous now, or so the pitch goes.

Cassidy Williams recently [wrote about this](https://cassidoo.co/post/vibe-coding-yawn/), and I can't agree more: there's no joy in vibe coding. There's no "I figured it out" feeling. It's just... there.

I've vibe coded a few tools for work to solve problems quickly. No argument, it gets the job done. Claude Code, Codex, and the rest are faster at generating code than any human. You can build in hours what used to take weeks.

But what we're missing is the understanding. The "I made tradeoffs A, B, and C because of these specific reasons." is something that I completely lost with vibe-coded applications. To make it worse, when I have Claude build an application from start to finish, the resulting code becomes write-only. It’s much faster to have Claude fix any potential issues than me trying to learn the code base.

In practice, I use code suggestions happily, but I've never let Claude Code build a feature on a production application. Monitoring whether it follows established patterns and style guides takes more effort than just building it myself. And building it myself is more fulfilling anyway.

So I don't buy the premise that AI should "do the boring work." What AI companies label as boring is actually where the meaning lives.
Of course an agent writes code faster. But speed isn't what we're measuring. We're measuring how well we can maintain a codebase for years. 

When we outsource the thinking, we hand our understanding to a black box. Then, suddenly, nobody knows why the database was designed this way. It's a coin flip whether the system will scale and where it'll break. 
The quality of what we ship suffers. We stop knowing how our own systems behave.