# Curl founder questions whether new Rust rewrites can match the original project’s pace
Two projects have recently attempted to recreate curl and libcurl in Rust, prompting curl founder Daniel Stenberg to argue that producing a codebase is only the beginning of the work required to compete with the long-running internet-transfer project.
Stenberg identified urlx, described by its creators as a memory-safe implementation built from scratch, and a second project presented simply as curl rewritten in Rust. The latter says it spent $8,600 on the effort. Stenberg counted about 340,000 lines of Rust generated from curl’s roughly 185,000 lines of C and said the conversion was completed over a little more than three weeks with extensive use of coding agents.
That speed is possible partly because curl already provides an unusually rich target, Stenberg wrote. Its source is public, its interfaces are broadly documented, many applications demonstrate how the API is used, and thousands of tests define expected behavior. Those assets allow automated systems to translate and repeatedly test an implementation against a mature specification.
The harder question is what happens after the initial rewrite. During the same period, the established curl project merged more than 300 commits and made slightly more than 200 bug fixes, according to Stenberg. He said the project averages about 11 commits a day. A compatible alternative would need to decide whether to follow that work continuously, diverge in another direction or risk falling behind on fixes and behavior.
Stenberg also challenged the idea that changing languages would remove most security defects. Of 81 curl vulnerabilities confirmed during 2026, he estimated 14 probably would have been prevented by a memory-safe language such as Rust. That represents 17.3 percent, leaving 67 issues whose underlying mistakes were not primarily about memory safety. The figures are Stenberg’s assessment rather than an independent audit of either Rust project.
His broader argument is that curl’s value lies in accumulated decisions, feedback, contributors and long-term interface stability as much as in its implementation language. A new clone may reproduce the software at a moment in time, but decades of compatible API and ABI behavior can only be demonstrated through continued operation.
Stenberg has not spoken with the teams behind the two rewrites and does not know whether they intend to develop production alternatives or proofs of concept. He nevertheless described their emergence as technically interesting and potentially useful. Attention to the problem could produce new ideas for internet-transfer software, even if neither project ultimately replaces curl.
The episode illustrates a growing distinction in AI-assisted development: agents can compress the time needed to generate and test a large body of code, while maintenance, governance and user trust still develop through repeated decisions over time.



