Maybe emphasize that more? In your example it just figures things out without iterating. How is this different/better than the Github copilot agnent mode: https://github.blog/news-insights/product-news/github-copilo...
Copilot sucks in general in my experience - but decent at finding relevance that's about it. I haven't tried this agent mode - will look into it. There are many AI agents so far seems for AI software development I'm looking towards: https://www.swebench.com/ for benchmarking - RA.AID has yet to benchmark as its quite expensive. But that is a good point I can highlight that difference thank you.