Who Wrote This?

A year from now, someone is going to look at a commit in one of my repos and want to know who actually wrote it. There’s a good chance that someone is me, and a good chance the honest answer is “I’m not sure.” Me, an agent, a different agent I handed it to, some combination, lightly edited. By the time the question gets asked, the context that could have answered it is gone.

People tend to treat this as a deep question, the kind you can’t really settle. If an AI wrote it and I directed it and I approved it, who’s the author? That’s a real question, and an interesting one, the kind I’d happily argue about for an afternoon. It just isn’t the question that helps me when I’m staring at the commit and need an answer.

I have had fun with that fuzziness before. I once published a report under the Open Futures Inquiry Lab, a research outfit that does not exist, staffed by a team of human and AI roles I invented. It was a bit, and a good one. But a commit is not a bit, and “I made up the byline” is not something you want attached to production code.

Because as an operational matter, “what or who wrote this” isn’t a question about authorship. It’s a question about logging. Did you record, at the moment the work happened, what produced it? If you did, the answer is just sitting there. If you didn’t, no amount of reasoning about the nature of authorship will reconstruct it after the fact.

And logging is a solved kind of problem. Commits can be signed, so you know which actor stood behind a change, and that actor can be a person or an agent as long as each has an identity. That much is just logging: what happened, and what produced it. Who authored it, and how much of it, is the next question, and it has a name: attribution. A convention, written down and followed, can mark what was AI-generated and at what level of human involvement, the same way you’d tag anything else you want to query later. None of it requires settling what authorship “means.” It requires deciding, before the fact, to write down what happened.

I’ve gone further on the convention side of this elsewhere, because attribution turns out to matter well beyond commits (disclosure isn’t the same as auditability, which is the actual argument). But the first move is the reframe. Stop asking who the author is. Start asking whether you wrote down what happened, because that’s the version of the question you can actually answer.