Yeah -- another vote for 2.5. My client used Claude on a script of mine that I was having problems with, and Claude pointed out a few issues -- so it was definitely useful as a second pair of eyes.

What it wasn't useful for was identifying that the original performance issue was the single-threaded approach to making web requests. Running these requests in parallel was the solution I needed. (It suggested doing a fetchall from the database instead of doing these queries on an as-needed basic -- this didn't improve performance, because that's not where the inefficiencies lay.)