The two sides of vibe coding.

Vibe coding tore down the wall between an idea and working software. The trouble is the wall was also a filter — and now anyone can build anything, the hard question isn't can you, it's should you.

Last time I argued the wall between an idea and a working tool had finally come down. This is the part where I admit that's not entirely good news.

Unless you've been living under a digital rock, you've heard the term. At its simplest, vibe coding means building software without knowing how to code — usually by speaking plain language into a tool like Claude Code or OpenAI's Codex. It takes a little more than that, and some technical sense, but that's the gist.

The upside is obvious: it's a democratization of who gets to make software. Anyone with a small budget and an internet connection can now build a tool — for themselves, or for everyone.

The flip side is exactly the same sentence. Anyone can now build a tool.

The avalanche

It's become so simple, and depending how you do it so cheap, that people are now building every last brain fart simply because they can. The numbers already show it: GitHub logged close to a billion commits in 2025 — 986 million, up 25% on the year and the biggest year in its history, with a record ~100 million in August alone. And that's just the floor. The slice that's actually AI-written is climbing far faster: commits from Claude Code alone went from essentially zero to around 135,000 a day between March 2025 and February 2026 — genuinely exponential, and accounting for roughly 4% of all commits on the platform.

I'm certain some astonishing things will come out of this — the kind that make you wonder why nobody built them sooner. But I'm equally braced for the avalanche of throwaway software that floats around the internet, half-finished, never really used.

I'm not immune. I have ideas all day, and every tool I touch I instinctively think: could I do this better? Could this be improved? Frankly, I do it with my own products, too.

A filter, not a faucet

I'm keenly aware that being able to build something doesn't mean I should. I haven't found the perfect rule yet, but I'm trying to settle on some rational filter an idea has to pass before it earns the right to exist. Sleep on it for a week, and build it only if I still want it? Get genuine interest from three people first? I don't know yet.

Some things I'll build just for the fun of it — that's allowed. But if I let myself vibe code every idea I have, I'm actively contributing to a world drowning in near-identical tools, each a hair different from the last, where simply choosing a piece of software becomes its own miserable chore.

So: vibe coding, yes. But everyone — myself first — let's hold ourselves to some minimal standard for the things we put out into the world.