Over the last year, I've been experimenting with LLMs almost every day, and I think the biggest shift isn't that models are getting smarter. It's that they're starting to do things instead of just answer questions. A few months ago I was mostly using AI to generate code, summarize docs, or brainstorm ideas. Now I'm finding myself building workflows where the AI plans tasks, calls tools, writes code, debugs itself, and completes work with minimal intervention. It feels like we're moving away ...