12. Models, Harnesses, and Tools
This week …
What a Large Language Model is, and what it is not. We separate the model from the harness that runs it, look at how tool use lets a model check and improve its own output, and compare subscription-based and API-based access. Then we look at R packages that connect R to LLMs, in particular btw, which lets a coding agent like Claude Code read R documentation, inspect your data, and run code in your R session.
Required Reading
These short posts by Hadley Wickham cover the main ideas:
- “What is an agent?”
- “Your LLM can’t math (and that’s ok)”
- “A coding agent is six functions in a trenchcoat”
- “Auto mode knows it’s dangerous (and does it anyway)”
Also look at the documentation for the btw and ellmer packages.