welcome
Thank you for purchasing the MEAP of Self-Improving Agents. To get the most from this book, you should be comfortable in Python and have called an LLM API before; if you have built even one simple agent or RAG application, you have all the background you need. There is no fine-tuning and no training math ahead, which is the point of self-improvement.
In the early 2000s, Jürgen Schmidhuber imagined the Gödel machine, a program that rewrites its own code the moment it can prove the rewrite is an improvement. The dream stalled on the word prove, and for twenty years self-improvement stayed theory. What changed is that the field stopped demanding proof and started measuring, and that one relaxation turned self-improvement into engineering you can ship.
I wrote this book because I kept watching teams build capable agents and then improve them by hand, tweaking and testing until the demo worked and hoping nothing else broke. Every one of those tweaks is a guess without a measurement, and the fixes could be seen rotting silently as models, data, and users' questions and expectations shifted. I realized there was a better loop, simpler than first thought, and it consisted of just: act, measure the gap, search for a better version, and gate what ships.