While AI agents have spent the past three weeks rewriting curl into Rust — 185,000 lines of C converted into 340,000 lines of Rust, a figure Daniel Stenberg says he counted himself — the actual curl project has quietly merged more than 300 commits, shipped a little over 200 bugfixes and fixed several vulnerabilities. The founder of the three-decade-old library that ships in everything from phones to cars to servers wants to know whether the clones can keep up.
During the time these agents converted 185,000 lines of C code into 340,000 lines of Rust code (I counted), we merged over 300 commits in the actual curl project, Stenberg wrote in a November 18 blog post titled Actual curl rewrites in Rust. We actually improved curl during these a little more than three weeks. A little over 200 bugfixes actually. 200 bugfixes that none of those rust rewrites have. We also fixed a few vulnerabilities. In another month we have done another 200 and maybe fixed a few more vulnerabilities. Will these clones keep up?
The post is a pointed reality check for one of the AI era's favorite genres of demo: the autonomous agent that rewrites a beloved open-source project in a modern language. Several independent projects have published curl-to-Rust conversions in recent weeks, generated largely by coding agents, and the line counts are genuinely impressive — the outputs roughly double in size as idiomatic Rust replaces pointer-heavy C. But Stenberg's arithmetic highlights what those demos omit: a rewrite is a snapshot, while the original is a moving target that keeps absorbing bug reports, security fixes and protocol changes every single day.
The timing is not accidental. The Rust-versus-C debate around critical network code has been accelerating all year. Anthropic revealed in October that it rewrote its entire Messages API in Rust with Claude writing the code, Microsoft researchers published work on Rust code that is memory-safe yet still breaks the language's safety guarantees, and open-source tooling such as the rust-anything project has emerged specifically to let agents migrate TypeScript, Python, Go, Ruby, Java and C codebases to Rust while recording the original application's behavior as a regression harness.
Stenberg's numbers also underline a subtler point about maintenance economics: the value of software like curl lies less in its 185,000 lines than in thirty years of accumulated fixes for edge cases — proxies, exotic TLS stacks, ancient platforms, protocol quirks — most of which never appear in any benchmark. An agent that regenerates the codebase inherits none of that history, and that history is precisely what the 200 bugfixes merged in the same window represent.
None of this makes agent-driven rewrites pointless — as experiments in rigor they are useful, and the tooling being built around them, from behavior recording to differential testing, is likely to outlive the demos. But for a project whose code runs unattended across a meaningful fraction of the connected world, the bar is not compiling and passing tests. It is keeping pace with a maintenance treadmill that has not stopped once in three decades. So far, Stenberg's verdict stands: the original is still moving faster.
Comments (0)
Log in to join the discussion
Log InNo comments yet