Recursive

In January 2026, a computational geometry specialist with twenty-eight years of software development experience had a throwaway idea about building a support chatbot for a client. Nine hours later, he had a working prototype. Sixteen days later, he had a multi-tenant platform with five deployed agents, a paying customer, and a codebase that could onboard a new client from a single URL.

This book is the honest, commit-by-commit story of that sprint-what worked, what broke, and why the decisions that mattered most had nothing to do with AI. It covers security hardening on day two (not day two hundred), a knowledge base built from twelve years of domain expertise rather than a generic document upload, an infrastructure migration that cut hosting costs while improving isolation, and a self-improvement loop where the agents propose their own knowledge base edits for human review. It also covers the parts that didn't go well: signups that never responded, a shared API budget that took down every agent overnight, and the humbling gap between "I have a product" and "anyone knows it exists."

The tools were new. The judgment behind every architectural decision was not. If you're an experienced developer wondering what AI-assisted development actually looks like in practice-not the hype, not the demo, but the real engineering tradeoffs-this is that story.