Write prompts that get good R code

I set out to make a lesson on writing prompts that get good R code, but I found that prompting matters less than it used to. When I tried to get Claude to give me base R code, it kept returning the tidyverse-style code I prefer, so this lesson ended up being about that and about the cases where prompting still helps.

You'll learn:

• Why generic prompts like 'make me a bar chart in R' now tend to return ggplot2 code, including a possible explanation based on the amount of ggplot2 training data
• How to prompt an AI agent to find base R code in an existing script and swap it for tidyverse-friendly alternatives
• How to state a specific package preference, such as fs for files and directories, so the code matches your style
• How standing instructions in a CLAUDE.md or AGENTS.md file can hold your coding preferences, such as the native pipe or ggplot2

Tools mentioned: Claude, ggplot2, tidyverse, fs, readr, CLAUDE.md, AGENTS.md

Key moments:

• 00:00 Lesson turns out different than planned
• 00:24 Livestream attempt to get base R code from Claude
• 01:02 CSV reading and combining also comes back tidyverse
• 01:18 LinkedIn replies from Jan and Gina Reynolds
• 02:00 Prompting still helps for specific preferences
• 02:34 Spotting base R dir.create in Claude's code
• 02:50 Prompt to replace base R with tidyverse-friendly packages
• 03:04 Claude swaps dir.create for fs::dir_create
• 03:44 Other preferences and CLAUDE.md or AGENTS.md