Its a lot more fun when you start playing around with finetuning different models, learning basic differences in their architecture and co.

You can experiment with different agent harnesses (pi, hermes, opencode, goose, ..) do experiemnets with gastown and paperclip ai.

I also get a lot of fun at experimenting with ideas i have and had were the focus is not the pure programming language but the application.

But i do think and we will see, as soon as there are clear winners of agentic platforms, we will finetune and adjust these agents.