This is a linkpost for https://kiankyars.github.io/machine_learning/2026/02/10/regex.html
At the beginning you say that there were 14 sessions but later (and in this post) say 1,273 sessions. The token counts you give seem to match a smaller number of sessions, so I'm wondering what the 1,273 number is?
Yeah. Sorry for the confusion. In my LinkedIn and X post I mentioned that the twelve hundred seventy three includes all of the sessions which were rate limited and that's also in the appendix. I only added that number in the title so that it would be flashier. In reality there are only 14 sessions where Claude was not rate limited.
I show two parallel Opus 4.6 agents build a fully functioning Rust regex engine after ~1,273 Claude sessions and 1.7K lines of code.
As I do not have the resources to build a c-compiler like is done in the original work by Carlini, the next task I am tackling is an embedded SQLite-like database engine in rust, which I predict to be 1 OOM larger in terms of LOC.