As Simon Willison notes, LLM is beginning to look 'very agent-shaped.' This is the key takeaway from his update. The tool's pivot from a clean API wrapper to a system that manages reasoning traces, tool calls, and stateful conversations reflects a maturation of the underlying technology, but also a potential consolidation. In our view, the risk is that such tools could become overly complex for their original purpose, while the opportunity lies in providing a unified, model-agnostic interface that prevents vendor lock-in. The real test will be whether developers adopt these advanced features for production systems or if they remain niche tools for prototyping.