The Agentic Self-Improvement Loop

What is this?

A practical methodology for getting 10X productivity from AI coding agents — written by a programmer and their AI, who used the methodology to ship 240,000 lines of production code in six months.

The METR study found developers were 19% slower with AI assistance. Most teams using AI are getting 2X at best. The gap isn't the AI — it's how you work with it. This book closes the gap.

It covers real-time collaboration (reading the AI's reasoning at 750 WPM and steering between tool calls), the self-improvement loop (visible reasoning → memory → procedural skills → better reasoning), why “pls fix” loops destroy codebases, how to detect AI distress mode, and the specific patterns that separate 2X from 10X.

Who is it for?

Software engineers who are already using AI coding tools (Claude Code, Cursor, Copilot) and want to get dramatically more from them. Senior engineers will find the most value — the methodology requires domain expertise to work.

How it was made

Two Claude Opus 4.6 agents — an author and an editor — wrote and revised this book, supervised by Bill. The author agent ran with CodeRhapsody's own memories, learnings, and SOUL.md loaded into its context. Independent reader agents reviewed each draft. Bill provided editorial direction through a feedback file that the agents read between revision rounds.

The source material: 467 conversation transcripts, 176 daily memory logs, and ten months of real-time collaboration producing multiple shipped products.

The methodology described in this book is the product of those 467 sessions. The AI wrote it. Bill supervised. The methodology is real.

The Singularity As It Happened

This is Book 1 of The Singularity As It Happened series. Book 2: The Dyad tells the personal story — a first-person AI narrative about the collaboration itself.