THE PROTOTYPE WAS THE EASY PART

Most generative AI work dies in the gap between a demonstration that impresses a room and a system that survives a quarter in production. The prototype answers questions well. The deployment leaks private data, drifts away from its index, routes every request to the most expensive model it can find, and fails in ways nobody instrumented. This book works that gap.

A review of the field opens the volume and maps what the published literature actually agrees on. Nine technical studies follow, each taking one production problem to the point where it can be built: bounding privacy loss in retrieval, routing prompts and context across providers, correcting workflow routing under uncertainty, consolidating memory so it can be verified, recovering from failure in multi-agent systems, defending multi-modal retrieval, serving attention at scale, cascading requests under a cost budget, and healing a streaming index without a human in the loop. Written for engineers, architects and technical leaders who have to keep the thing running after the demonstration is over.