What the Structure Does Not Carry

The architecture diagram of every enterprise has always been missing one component.

Between the boxes and arrows stands an invisible layer of people, carrying what the integrations never could: what a figure means, who approved it, why an exception survives, how separate systems stay aligned. For forty years, that layer made the design look adequate - because it never stopped.

Agentic AI does not add a requirement. It withdraws the subsidy. As autonomous agents take their place at the crossings, the human carriers leave - and what the structure does not carry, nothing will.

Approaching the enterprise as an engineer rather than a software developer, Georg Zangl asks the question engineering asks first: what must be conserved? Six properties must survive every boundary that work crosses:

• Context - what a figure means, beyond its value
• Authority - who may act, and on whose say-so
• Accountability - who answers when the output is wrong
• Coordination - how separate systems stay aligned
• Coherence - whether work completes or fails cleanly
• Institutional memory - the reason a decision was made

What the Structure Does Not Carry states the law these properties obey, derives the enterprise operating infrastructure that satisfies it, and tests the whole argument against documented failures sourced to courts and regulators - Canada's Phoenix payroll, the Post Office Horizon scandal, Knight Capital's forty-five minutes, and more.

This is not a book about AI hype, and not a vendor pitch. It arms your scepticism rather than asking your agreement: every chapter ends with tests you can put to any platform, landscape, or vendor, and it closes with six custody rules to carry away.

Who it is for: CIOs, CTOs, enterprise architects, CISOs, and boards making decisions about agentic AI, governance, data sovereignty, and digital transformation in regulated industries - and anyone responsible for what happens when the people are removed from the loop.

Georg Zangl is a petroleum engineer who has built engineering and AI software since the 1990s, founding his first AI company in 1996. He now builds agentic-AI infrastructure for regulated enterprises. The principles in this book were extracted from that work - not derived at a whiteboard.

The enterprise is its properties. Everything else is equipment.