90% Failure Rate: Why GPT-4o Can't Optimize Code (And What We Built Instead)

GPT-4o fails 90% of the time at code optimization. Why zero-shot prompting doesn't work and how to build an agentic search workflow instead.