I tried Cursor as I was excited it used Tree Maps for RAG. I tried it for 3 different medium asks on a medium sized project (https://github.com/Kareadita/Kavita).
I was extremely disappointed. It recreated enum files with the same values, tons of changes, overall I rolled back.
I use copy/paste into ChatGPT/Claude a lot. I will stick with that. I control the scope, I control what I copy or how I take the solutions.
I stand by my belief that AI cannot work on medium to large codebases. I know we see vibe coding as a trend, but those are usually greenfield applications.
I was extremely disappointed. It recreated enum files with the same values, tons of changes, overall I rolled back.
I use copy/paste into ChatGPT/Claude a lot. I will stick with that. I control the scope, I control what I copy or how I take the solutions.
I stand by my belief that AI cannot work on medium to large codebases. I know we see vibe coding as a trend, but those are usually greenfield applications.