Blog
ralph-starter vs doing it manually
I tracked one full sprint week. 6 tasks manual, 6 with ralph-starter. 45 min vs 12 min per task. $1.87 total for the automated half.
I tried 5 AI coding agents with ralph-starter
I ran the same JWT auth task on Claude Code, Cursor, Codex CLI, and OpenCode. Real times, real costs, real results.
Ralph Wiggum technique explained in 2 minutes
The Ralph Wiggum technique is running AI coding agents in loops until done. Created by Geoffrey Huntley, now used by Claude Code and others.
Specs are the new code
A clear 10-line spec gets you a working PR in 2 loops. A vague one-liner wastes 5 loops and costs 3x more. The spec is the code now.
Prompt caching saved me $47 last month
Prompt caching gives 90% off input tokens after the first loop. My bill dropped from $109 to $62 last month without changing anything.
How I ship tasks from Linear every day with AI
My daily workflow with ralph-starter and Linear. Morning standup, label tickets ralph-ready, batch process while I work on the hard stuff.
Automating entire workflows with ralph-starter
ralph-starter runs Ralph Wiggum loops. Fetch a spec, run the AI agent, check tests/lint/build, feed errors back, repeat. Here is how it works and why I built it.
Building a full app from a Figma file in one command
Pointed ralph-starter at a 12-screen Figma dashboard file and had working React components before the weekend. 87 cents total.
ralph-starter + Claude Code: the full setup
Zero to your first automated PR with ralph-starter and Claude Code. Install, init, run, and get a working PR in under 2 minutes.
My first ralph loop: what actually happens
Walk-through of a real ralph-starter loop from start to PR. What happens at each step and why the loop design matters.
Why I built ralph-starter
I was copy-pasting between ChatGPT and my editor 20 times a day. So I wrote a bash script that did it for me. That script became ralph-starter.
Let your tests guide the AI
AI-generated code that looks perfect can blow up at runtime. Adding tests to the loop lets the agent catch and fix its own mistakes.
From spec to code in one command
Your specs already live in GitHub, Linear, or Notion. One command pulls them into ralph-starter and starts coding.
Why autonomous AI coding loops work
I spent months copy-pasting between ChatGPT and my editor before I realized the loop itself was the problem, not the AI.