Building a Self-Improving Product with the Ralph Loop

Create a cron job that runs daily (Ryan runs his at midnight) to collect data from your database, users, and analytics - the same data you'd look at in a marketing dashboard.

// Run daily at midnight
// Collect: user behavior, free trial data, feature usage, errors
// Pipe this data to Opus 4.5 or GPT-5.2

Have the AI analyze the collected data and generate a markdown report identifying the one thing you should fix. The report highlights holes in your funnel or opportunities for improvement.

Instead of manually acting on the report, have the system automatically generate a PRD (Product Requirements Document) from the identified opportunity. This is the key insight - don't be the bottleneck.

Create the full loop: Analysis → Report → PRD → Implementation → Review. The loop runs autonomously, creating PRs that implement the PRD recommendations. Named after Jeff Hunt's original concept, built on Kieran Klaassen's compound engineering.

Ryan created an open source repo for compound product that works across different agent tools (AMP, Claude Code, Cursor). The pattern is tool-agnostic.