Solo AI development with Claude Code

A map of one way to build software alone with Claude Code, from an idea to working code. It follows Anthropic’s published guidance as of November 2026, with a few practices of our own, marked as such.

Two versions

Quick start

For small, low-stakes projects: personal tools, prototypes, experiments.

Four stages: set up, describe and spec, build loop, review.
Session habits, common failures, a short file guide.
About 15 minutes to read.

Full guide

For software that other people will use, or that touches money, accounts or personal data.

Intent, spec and roadmap, then one build cycle per phase.
Tests and CI, mutation testing, hooks, security controls.
About an hour to read.

Switch between them at any time with the control at the top left.

How the map is laid out

The main flow runs left to right through the middle. Solid arrows show the order. Dashed arrows show where you loop back.
Regions above the flow explain why each step exists.
Regions below the flow show how: files, checks, prompts.
A yellow note and a worked conversation sit under each stage.
Each card shows its source: Anthropic guidance or Our recommendation.
Drag to pan. Pinch, or hold Ctrl or ⌘ and scroll, to zoom. Use “Go to” to jump.

Use this guide with Claude

Claude can read this whole guide and set up a project to match it. Download the guide, put the file in your project folder, and give Claude the prompt below. Andrej Karpathy shares his own projects this way: the data and a prompt, so each person can have Claude build their own version.

The Markdown file holds the same content as plain text. Claude reads it with fewer tokens.

Show this screen at start