‘Ralph Wiggum’ loop prompts Claude to vibe-clone commercial software for $10 an hour

Summary

Open-source developer Geoff Huntley created a simple bash loop — dubbed “Ralph” — that feeds an AI coding assistant’s output back into itself until it produces acceptable results. By driving Anthropic’s Claude Code in an automated, agentic loop (for example: while :; do cat PROMPT.md | claude-code ; done), Huntley says the system can produce high-quality software at very low cost (around US $10 of compute/SaaS per hour).

Huntley has used Ralph to build a ZX Spectrum tax app, reverse-engineer and clone an Atlassian product, and even design a deliberately odd new language called “Cursed.” He warns the technique changes the human-in-the-loop model: humans intervene later and less often, and startups could use such loops to clone commercial SaaS features cheaply and rapidly. Anthropic has taken notice and published a Ralph Wiggum plugin for Claude Code.

Key Points

“Ralph” is a persistent loop that repeatedly feeds AI output (including errors) back into the model until it converges on acceptable code.
Huntley used the method with Claude Code to clone and recreate commercial and open-source products, improving results by feeding docs and specs into the loop.
The approach can be very cheap — Huntley estimates roughly $10/hour in compute/SaaS — which is far below the cost of human developer labour.
Agentic loops shift when and how humans participate: less frequent intervention, more automation and iterative self-correction by the assistant.
There are broad implications: product features can be cloned quickly, potentially undercutting incumbent SaaS businesses and disrupting developer workflows and roles.

Author style

Punchy: this write-up highlights a potentially seismic change in software development and amplifies why the details matter — Huntley’s idea isn’t just clever tinkering, it’s a warning and a provocation about where coding and business models might head next.

Why should I read this?

Because it’s both fascinating and a bit creepy. If you care about how software gets made — or how your product might be copied for pocket change — this is a short, sharp briefing that saves you the time of digging through the original thread. Plus: there’s a neat one-liner loop you’ll want to know about.

Context and Relevance

Agentic AI and coding assistants are increasingly capable of producing production-grade outputs. Ralph illustrates a practical, low-cost way to automate iterative code generation and testing, which ties into wider trends: cheaper inference, tooling that reduces human touchpoints, and the commercial risk of rapid cloning. For startups, vendors and maintainers of open-source projects, the technique raises questions about competitive defence, licensing and the future of developer roles.

Source

Source: https://go.theregister.com/feed/www.theregister.com/2026/01/27/ralph_wiggum_claude_loops/