The book is written for industrial software architects and operational-technology engineers who must decide where a generative model may and may not be trusted, for AI researchers and product teams building agentic systems for high-stakes domains, and for graduate students seeking a compact, opinionated treatment of the subject.

A word on what the book does not claim. The architectures and evaluations presented here illustrate and stress-test the central argument; they align with established functional safety principles but do not constitute safety certification, and several of the evaluations are conducted in simulation or on prototypes rather than on production plants. Where this is the case it is stated explicitly, and the boundary between what has been demonstrated and what remains to be validated is drawn as clearly as the author could manage.

The work synthesized here grew out of a coordinated line of research, and it owes much to colleagues at Siemens and to collaborators in the academic community whose ideas, criticism, and patience shaped it. It builds on themes the author first developed in two earlier monographs on service-oriented crowdsourcing and social-network-based recommendation, and it is offered in the same spirit: a focused architectural argument, made precise enough to be useful and to be argued with.

Daniel Schall is the head of the distributed AI systems research group at Siemens AG Austria in Vienna. His main interests are industrial AI & vision systems, platforms, and applications that have real-world impact. He is interested in the intersection of edge computing and artificial intelligence and their embedding in industrial systems. He is the author of Service-Oriented Crowdsourcing: Architecture, Protocols and Algorithms (Springer, 2012).