I have mine too!

He's more of a colleague.

I wish you did :)

Must be something in cursor system prompt.

Is Claude proactively suggesting new optimizations?

The thing is that LLMs (especially reasoning models) can actually be very good at decomposing problems when given all of the relevant context. As LLMs get better at gathering context and using longer inputs more effectively programming with AI will evolve.

You should also try caching.