A general-purpose AI assistant such as ChatGPT or Claude is three separable layers working together: the model (the underlying LLM doing next-token prediction), the product (the chat application wrapping it: conversation storage, memory, file upload, the interface), and the tools the product can invoke on the model's behalf (web search, code execution, image generation, file readers). Treating these as one thing is the most common source of confusion about how these systems behave.