Hot take season says open source is dead because AI trained on public repos.

I don’t buy the funeral. I do buy the part where the old deal is broken.

I use open source all week. I also try to support it: issues, the odd sponsorship, publishing small things in public when they might help someone else. That sharing habit is a big reason the web is buildable. It’s also why the current moment feels awkward. I use AI coding tools. They help. They learned from public work. I’m in the loop I’m complaining about.

What actually changed

For a long time, “public” mostly meant: other humans can find this.

Now it also means: a crawler can train on it, an assistant can remix it, a company can sell the distilled version back by subscription.

That’s why people are angry. Not because a model can read code. Because work released under open licences got turned into a shop window.

You saw the Copilot era arguments: blender metaphors, leave-GitHub campaigns, a copyright case that mostly failed in court. The legal scoreboard and the cultural bruise weren’t the same thing. Licences never really encoded “no commercial model training”.

And AI didn’t invent this. Cloud vendors got there first. Redis, Elastic, HashiCorp, MongoDB and others already moved toward source available terms after “open” started meaning “please compete with us using our own code as a service”. AI just made it feel personal. Your dull utility and weekend library get vacuumed up, then rented back.

The response: fees and blocks

Maintainers and platforms are doing the predictable thing.

Cloudflare’s pay per crawl: allow, charge, or refuse. Automattic: block major AI crawlers by default on WordPress.com and Tumblr. Longer robots.txt files. Experiments like ai.txt that try to say crawl is not the same as train.

Good. Also dangerous.

If every useful README needs a tollbooth, juniors lose the midnight learning path. Small projects don’t have legal teams. Big publishers can negotiate. A lone maintainer has crossed fingers. “Protect the commons” can become “only the well resourced stay findable”.

That’s my actual worry. Not that GitHub goes dark. That the default posture becomes locked and suspicious, and shipping something helpful in public starts to feel naive.

Practical fork

Readable ≠ open. Marketing will keep blurring that. Don’t let it.

If you publish: be clearer about what you’re offering, especially around training.

If you use OSS: put something back.

If you build AI on public work: a real relationship beats a thanks page.

If you run a site: decide whether crawlers are guests, customers, or trespassers, and own it.

I still lean publish. I get the gate.

I almost didn’t push a tiny public repo recently for exactly this reason. Old instinct: someone else might need it. New instinct: this becomes training data, then a paid feature. I published anyway. Not everyone will.

Which downside can you stand?

I wrote the longer version, with more of the trail, on my site:

Open source isn’t dead. The free lunch is.

The code can stay open. The lunch was never going to stay free forever.