Scanning
Running Multi-Agent Systems in Production
Demos are easy. Production is a different discipline.
Almost everything written about multi-agent systems describes getting one to work. Very little describes keeping a fleet of them running against real traffic, real data, and real consequences — where a retry storm costs money, a silent degradation costs trust, and a single unbounded loop can run all night.
This is the engineering textbook for that problem: orchestration topologies, state and memory boundaries, cost and token governance, failure modes and their signatures, observability for non-deterministic systems, evaluation that actually catches regressions, and the operational practices that keep an agent fleet accountable to the business paying for it.
Structured as a textbook with worked examples and exercises rather than a narrative.
Tim de Vallée
Tim de Vallée built the US operation for IMDb in the mid-1990s — from a hard disk shipped to a volunteer in Wisconsin to the first internet ad campaign a Hollywood studio ever bought. He now builds intelligence tools at INTLO Labs, the product house of Digital Boutique AI, drawing on thirty years of shipping things.
More about the author →