llm 0.35

Your team needs to know what AI models are running, where, and on what data. The constant arrival of fresh models and simple ways to access them means people will try them out. This is how new capabilities are found. But it also means you quickly lose track of what is being used. This creates a hidden cost in unmanaged data and a lack of clear audit trails, not just the API fees.

Simon Willison's llm tool shows this trend. He announced version 0.35, adding support for OpenAI's gpt-6-astra model. Tools like this make it simple for anyone to pull in the latest AI from their command line. The sponsor message on Willison's page highlights the problem: "Shadow AI is the new shadow IT." This points to the real challenge of governing these fast-moving capabilities without a central plan.