curl author Daniel Stenberg has revealed a set of numbers worth noting: over just three weeks, a coding agent rewrote roughly 185,000 lines of C into 340,000 lines of Rust. During the same period, curl merged more than 300 commits and actually fixed around 200 defects not covered by the rewrite. Our judgment is that AI programming has moved from demo code into real production projects, but “the machine finished writing it” does not mean “the software has become better.”
What This Is
A coding agent—an AI system capable of autonomously breaking down tasks, modifying code, and invoking tools to complete development work—was used to migrate part of curl’s C codebase to Rust. Stenberg also emphasized that the project did not stall during the migration; it continued to receive bug fixes and security patches. This means AI is more like a fast-moving execution team than a single replacement for maintainers.
Industry View
These figures support a pragmatic conclusion: AI’s value for large legacy systems lies first in time-intensive work such as migration, refactoring, and batch processing. It can also free people to focus on harder architectural decisions and security reviews. Yet Stenberg himself raises the key question: can the automatically rewritten Rust version keep pace with the original project’s continuing stream of fixes?
The risks are equally clear. An increase from 185,000 to 340,000 lines of code does not automatically imply better quality; automatically generated code may amplify hidden defects and introduce inconsistent implementations. Without adequate review, testing, and rollback mechanisms, the larger the migration, the higher the potential maintenance cost. The curl case is valuable precisely because the project keeps releasing real fixes rather than merely presenting generated output.
Impact on Regular People
For enterprise IT: Companies can prioritize AI for legacy-system migration, test completion, and documentation cleanup, but human review, security testing, and version governance must receive equal attention.
For individual careers: Programmers’ work may shift from writing code line by line toward defining requirements, reviewing results, and owning quality. The ability to collaborate effectively with AI will have a more direct impact on productivity.
For the consumer market: Faster refactoring may reduce software maintenance costs and accelerate feature iteration, but consumers will ultimately judge the result by stability and security—not by who wrote the code.