I wonder if for anyone it seems like the more agentic LLMs get, the more difficult some things have gotten or going a certain route more often in responses, compared to running a similar task on - a local model?