I like this story. How did you verify the output? How big is the codebase? Why it took 18 hours? Could you implement it with a small local agent and breaking the task down yourself in two days (i know it sounds like a loaded question, it is not).
I think the rewrites are the main story for llms in code (hot take). Writing greenfield code at the seams also something which might work well.
It's a fairly large code base split across 3 repos.
The good thing was that it is fairly easy to verify: we have a working (but slow) version that uses Spark, with lots of existing unit tests.
We verified by using those unit tests as well as running our end-to-end process in the Spark and Pandas version and verifying the two databases were within the differential-privacy noise bands of each other.
I think the rewrites are the main story for llms in code (hot take). Writing greenfield code at the seams also something which might work well.