The Larder Lesson
The foundation essay makes the case for AI harnesses abstractly. The abstract version does not land for people who have not tried one. This lesson is the applied version.
Larder in Claude Code builds a small harness around a mundane worked example: planning a week of dinners. Recipes, a pantry, a preferences file, and a terminal.
Two kinds of work
Meal planning is two kinds of work in one task. One is unstructured: reading recipe pages that were written for humans, noticing that a constraint from last week contradicts this week’s plan, deciding whether Saturday feels right for a big cook. The other is structured: subtracting the pantry from the week’s ingredients, scaling servings, grouping the shopping list by aisle.
An LLM is cheap at the unstructured work. A small script is cheap at the structured work. You are the one who decides which constraints matter and when the system is wrong. A harness is the small set of files in a folder that gives each piece of a task to the tool suited to it and leaves the judgment with you.
Who I made this for
People who use computers every day but do not write code for a living. These lessons are the concrete answer to “what do you actually do with Claude?”, because the abstract answer is hard to follow if you have never built one. What transfers is the habit: correct the file instead of the answer, and the corrections you make in one week compound into the system you run the next.
What is next
- Larder in Claude Code for the applied build: a folder of files you can read and edit yourself.
- AI harnesses are for everyone for the conceptual foundation.
Pick something you actually want done this week. Not the thing I picked. Yours.