I posted the other day about basepage, the local-first tool I'm building that lets you point an agent at a folder of files and have it build and publish a site. Someone replied with a flat objection: AI is the opposite of the indieweb.
His case was clean. The indieweb is about ownership, POSSE, choice, and openness. AI, he said, is the exact opposite: dependency, deskilling, no choice, and closedness. And then the sharpest version of it: indie is a thing for people, not agents. You can't be indie with an LLM.
I've been chewing on it since, because he's half right, and the half he's right about isn't the half people usually argue over.
Start with the part I think is wrong. The claim that a site built with an agent can't be indie doesn't survive a close look. basepage isn't a new black box. It's an assemblage of ordinary indieweb tooling, markdown files, git, GitHub Pages, RSS, the URL, packaged so that an agent can operate it as well as a person can. The thing that comes out the other end is plain files on open infrastructure. You can read them, edit them, host them anywhere, and walk away with them whenever you want. That is as indie as a site gets, and none of it changes based on whether a human or an agent typed the commands. The principles attach to the artifact and the stack underneath it, not to whose hands were on the keyboard.
So the site is fine. But that was only one of the two things he was actually saying, and it's the weaker one.
The stronger objection isn't about the site at all. It's about the person. Dependency. Deskilling. Indie is for people, not agents. Even if the output is portable and open, the worry goes, you have handed the doing of it to a model you don't own and can't inspect, and that is a different thing from making it yourself. My assemblage argument doesn't touch this, because it isn't a claim about files. It's a claim about autonomy.
And here it gets harder to wave away. Is depending on an LLM really different from depending on 11ty, or a static host, or a framework, or a CDN? Every indieweb site already sits on a tall stack of tools the maker didn't build and mostly couldn't rebuild from scratch. If dependency disqualifies you, then none of us were ever indie, and the agent is just one more layer on a pile we already accepted.
But I don't think that fully escapes it, and this is where I'll give him the point. The difference is ownability. 11ty is open and sitting on my disk. I can read it, fork it, and keep running it for as long as I like. A frontier model is none of those things. It's centralized, opaque, controlled by a handful of labs, and it can change or vanish under me without my say. That is a real difference, and pretending it isn't would be dishonest.
So here is where I land. The output can be fully indie no matter how it was made. The practice is where the real question lives, and the answer isn't "AI is fine" or "AI is the opposite of indie." It's a choice about how you hold the tool. Treat the agent as an operator of primitives you understand and could fall back to, not as a replacement for understanding them. Keep the formats open and the exit cheap. Do that, and even a closed model becomes one more swappable layer instead of a dependency you can't leave. The same logic probably runs further than I've taken it, toward local and open models I haven't really tried yet, but it's a direction I can already see the sense in.
Because I don't think indie ever meant doing everything yourself from raw materials. It meant being able to leave. As long as I can take my files and rebuild my site without asking anyone's permission, a lab's included, I'm still indie.