Exactly a week ago, writing about curl turning 30, I said that what I liked most about the story was that curl has never been rewritten. Well, what do you know: seven days later Daniel Stenberg himself explains that somebody has done it… though not him, and not in C. On his blog he talks about at least two rewrites of curl in Rust that have appeared this autumn, made with AI agents.
The rewrites
The first one is urlx, which presents itself as a “memory-safe, from-scratch reimplementation of curl and libcurl”. The second, which appeared this month, is simply called curl, described as “a rewrite of curl and libcurl in Rust”. Stenberg hasn’t talked to any of their authors; he found out through links people kept sending him, and he suspects there are more.
He gives a few numbers about the second one, and they’re worth putting side by side:
| What | How much |
|---|---|
| Cost claimed by its author | about 8,600 dollars |
| Duration | a little over three weeks |
| C code to start with | about 185,000 lines |
| Resulting Rust code | about 340,000 lines (he counted) |
| Commits to the real curl in those same weeks | over 300 |
| Of which bug fixes | a little over 200, several of them security fixes |
The cost comes with a caveat, as the comments point out: it’s the estimated equivalent at API prices, because the work was done on a subscription. Either way, for the price of a good laptop you get 340,000 lines of Rust that do (more or less) the same as curl. Five years ago this would have been science fiction.
Writing the code was never the problem
And here’s what I find most interesting, because Stenberg’s reaction isn’t that of someone who feels threatened. He sums it up himself: he’s “amused. A little puzzled. Fascinated”. His argument fits in one of his sentences: “writing code once was never the issue”.
What makes curl valuable isn’t the 185,000 lines, but everything around them: years of iterations, the test suite, the community, and decades of API and ABI stability. “curl is the result of time”, he says. And time can’t be copied: while the agents were translating, the original project kept fixing bugs at a rate of around 11 commits a day. Each of those fixes is something the rewrite doesn’t have, unless somebody ports it.
It’s like taking cuttings from an old grapevine. You can take the cuttings in an afternoon and they have the same DNA, but the vine has spent decades putting down roots, surviving droughts and producing grapes every year. Whether the cutting ends up becoming a vine doesn’t depend on how quickly you took it, but on who’s going to water it for the next twenty years.
The memory safety argument
The usual reason for rewriting in Rust is memory safety, and Stenberg has the data at hand to answer. Of the 81 vulnerabilities confirmed in curl in 2026, only 14 (17.3%) come from mistakes that a memory-safe language would probably have prevented. A Rust version would have had the other 67.
In the comments somebody argues that Rust brings more than memory safety, and he answers honestly: what would have happened in another language can’t be known, but past C vulnerabilities are public and anybody can check them. It isn’t an argument against Rust, mind you, but against the idea that changing language fixes a project’s problems: most bugs are logic bugs, and those get rewritten just as well in any language.
The intern who translates very fast
This fits quite well with what I think about AI, which I wrote about in the watercolour article: it’s a great intern for clear tasks, and “translate this code from C to Rust” is a very clear task. What the intern doesn’t do is stay for twenty years maintaining what it translated, answer users, review security reports (which, as I mentioned last week, in curl are already largely AI-generated noise) or decide what not to do.
Stenberg wishes the authors of the rewrites luck, and so do I. Maybe one of them ends up with its own community and a long life, and that would be great. But for now they’re proofs of concept of something we already knew: code has become cheap, and what’s still expensive is everything else.




