chapter two

2 The evolving role of the architect

This chapter covers

Where traditional architecture frameworks need to be extended
Six architectural principles for governing delegated autonomy
The architect as guardian of trust
Designing teams for human–agent collaboration

Enterprise architecture has always been about deciding who is allowed to decide, under what conditions, and with what consequences. For years, we exercised that responsibility at design time, reviewing architectures, approving roadmaps, and trusting that systems would behave as designed once in production.

Agentic systems break that model. Decision-making no longer stops at approval; it continues in production, at machine speed, often between governance cycles. The architect's job cannot end at sign-off. We must design for supervision as well as specification, making explicit where authority has been delegated, how far that authority extends, and how we will know when to question that delegation.

2.1 Principles for the agentic enterprise

2.1.1 Principle 1: Autonomy must be bounded

2.1.2 Autonomy must remain attributable

2.1.3 Autonomy must remain observable

2.1.4 Principle 4: Autonomy must remain governable

2.1.5 Principle 5: Autonomy must transfer safely when it reaches its limits

2.1.6 Principle 6: Autonomy must be continuously realigned

2.1.7 Making the principles usable in practice

2.2 Architects as Guardians of Trust

2.2.1 From Blueprint Maker to Resilience Engineer

2.2.2 Designing for human oversight

2.2.3 Skills for the next decade: Simulation, Policy-as-Code, Orchestration

2.3 Organizational design for augmented teams

2.3.1 Teaming models for humans and agents

2.3.2 The team–agent contract

2.3.3 Antipattern: Agentic silos (Conway’s law at machine speed)

2.4 Summary