2026-06-25 Thu
- Filed
- An explanation of what Claude/Code is, framed as "the story of how we got here" rather than a definition — given to a colleague who asked whether Claude Code is an AI/LLM.
- The history of LLMs is largely the history of giving the LLM new tools it can "choose" to use.
- Nov 2022: ChatGPT was an LLM with a single tool, "respond" — take a chat history plus a new user message, return one AI message. This single tool defined the chatbot era.
- Later: ChatGPT "gave" the LLM tools tied to the Canvas in the web UI, letting it choose when to create or update a document shown to the right of the chat.
- Later: ChatGPT "gave" the LLM a Code Interpreter tool (later Advanced Data Analysis), letting it choose to run Python and inspect the output when that would help.
- Claude/Code and Codex are the latest evolution: they "give" the LLM tools that act on your computer — read a file, create a file, modify a file.
- This class of tools is an AI/Agent/Agentic harness: the LLM runs in a loop, calling tool after tool. Most of what it does is not replying to you but invoking tools that correspond to things the operating system already knows how to do.
- Whether in Claude Desktop, the VS Code extension, or the terminal (or the equivalents from other companies), "Claude Code" refers to an application that lets an LLM ask the operating system what it can do, then use it to take actions.