> what prompted this post? well just tried to work with gpt5 and gemini pro That's the problem. GPT5, at least in Cursor, doesn't work for coding. It burns tok

> what prompted this post? well just tried to work with gpt5 and gemini pro That's the problem. GPT5, at least in Cursor, doesn't work for coding. It burns tokens and does actually nothing my in experience. Claude 3.5, 4 and 4.5, on the other hand, are pretty solid and make lots of forward progress with minimal instruction. It takes iteration, some skill, some critical thinking, and some hand coding! Yes, LLMs forget things and do random things sometimes, but for me it's a big boost.