At EuroPLoP 2026 we convened a focus group of 22 practitioners from industry and academia to examine how AI agents are reshaping software architecture work. Participants discussed existing practices, trust and validation methods, the limits of AI autonomy, governance hurdles, and educational implications.
The consensus was that architectural decision‑making, accountability, and the definition of architectural guardrails remain fundamentally human tasks. An emergent notion called "governance engineering" describes the discipline of building the system that oversees AI‑assisted creation, comprising validation mechanisms, knowledge layers, and organization‑specific standards.
Attendees agreed that criticality—defined as the product of uncertainty and change‑cost—is a universal metric for calibrating human oversight. A further concern is cognitive debt: the gradual erosion of human understanding when AI‑generated decisions are accepted without deep intellectual engagement.
The report outlines future work such as quantifying criticality, designing auditable governance‑engineering frameworks, and integrating AI‑augmented architecture thinking into curricula.
Review