Since I started using all flavors of Gemini 2.0 I haven't been back on Cursor Gemini Exp 1206 and Gemini 2.0 Thinking Model exceed o1 in my experience and best

Since I started using all flavors of Gemini 2.0 I haven't been back on Cursor Gemini Exp 1206 and Gemini 2.0 Thinking Model exceed o1 in my experience and best part is that its free with insane context token size. For agentic code experience RooCline is quite good but I actually do most of my work in ai studio and like to create the folders and files manually. I almost think agentic code generation is the wrong step forward. Without knowledge transfer/learning involved during the code generation phase, you end up cornering yourself into unfixable/unknowable bugs. You can see the examples on the linked site are quite simple. These are hardly real world use cases. Most of us aren't generating things from scratch but rather to gain understanding of systems being worked on by multiple people and very carefully inspecting and undertanding the output from GPT This is why a lot of tools like Bolt, Lovable, Cursor, Windsurf will be toys aimed at people looking to use codegen as a toy rather than a tool. It will serve you better if you look at AI codegen as a knowledge transfer/explorer tool and an AI agent isn't wise imho.