I really liked the evals at the end for building a simple benchmark to test various coding agents. I get asked all the time how do do this: magazine.sebastianraschka.com/p/using-local-coding-agents