Creating with Claude
A working guide to co-creating a product, a business, or any expansive thing with Claude, where a clean, refined concept is the source and everything you ship is an artifact of it.
The mission is to teach you to build regular, deterministic software with Claude. Claude is the builder, the reviewer, and where useful a scheduled operator; those are agents, and you’ll use them freely. The line this course draws is about where the LLM sits: the product you ship runs deterministically, reaching for LLM judgment only in narrow, internal, human-gated places. Building a product with an LLM at its center for end users is its own discipline (sanitizing inputs and outputs, prompt iteration, evals, resisting misuse) and is out of scope here, though much of the course applies if you go there.
A note on how this was made: this course is heavily LLM-generated, written with Claude and edited by me. The point is not that every detail is perfectly correct or current. The point is the topics: an opinionated backbone of what I think is worth knowing about creating with AI. Read it with an LLM alongside, so you can ask questions, check specifics against current docs, and go deeper wherever you want.
Read in order the first time. Skip around after.
Before you begin
I: Getting set up
II: The big idea
- A refined concept is the product
- Markdown as the medium
- Everything through Claude
- Stay close to the primitives