Claude Code Changes How I work (Part 3): Creating Devil’s Advocate Agents for Tough Problems

I am going to be throughout 2026 returning to my old paywalling method where I will each post randomly flip a coin 3 times and best two out of three determines if I paywall it. I run the code here at google Colab each time. This time it was heads two out of three so this is a paywalled post.

This is a third post in a series I’m doing trying to learn about using Claude Code of empirical research. In this one I talk about a problem I ran into for scraping an online academic genealogy site. I experimented with an idea to try and get Claude Code to better anticipate the problems with the strategy we were taking because I felt like we were spinning our wheels. It was one of those LLM things of over-confidence in diagnosing a problem, quickly solving it, but a delay in me understanding the problem, the solution and realizing the error. I call it “creating a devil’s advocate”.

Thank you everyone for your support of the substack which is a labor of love. I much appreciate your support of it and hope you find this helpful.