Simon Willison demonstrates using LLM in shell script shebangs to generate content and run tools

LLM (an AI tool that generates text) can be invoked directly from a script's shebang line using the pattern `#!/usr/bin/env -S llm -f`, allowing a text file to function as an executable prompt.

The approach supports tool calls via the `-T` option and YAML template execution via the `-t` option, which can define custom Python functions like `add()` and `multiply()` for the LLM to invoke during execution.

Example use case: a script can accept a math expression like `'what is 2344 * 5252 + 134'` and output both the tool calls made and the final result (`12,310,822`).