Artificial intelligence dominated the discussion at NXT BLD 2026, but Campbell Yule argues that the term agent is being applied too broadly across AEC technology. He identifies four different categories being presented as AI agents and explains why distinguishing them matters, tells AEC Magazine.
The first category is the familiar parametric tool: defined inputs pass through deterministic logic to produce defined outputs. These systems are valuable because architects can inspect, question, and override their logic. Calling them agents can make transparent tools appear more autonomous and opaque than they really are.
The second category uses natural-language prompts as interfaces to existing capabilities. Prompts can simplify rendering, model searches, and specification drafting. However, Yule sees a major weakness: successful prompt-based workflows are rarely captured, versioned, governed, and reused. A useful result produced once remains a demonstration unless firms can turn the process behind it into a repeatable tool.
The third category involves software developed rapidly with AI. A.Engineer, for example, enables teams to create calculation engines, structural utilities, and analytical tools much faster than before. As development becomes cheaper, verification, validation, approval, monitoring, and retirement become more important. In regulated AEC work, the governance surrounding a tool may matter more than calling it an agent.
The fourth category comes closest to genuine agency: systems that accumulate knowledge across projects and apply it to future work while keeping their reasoning open to professional scrutiny. Such systems could transform successful prompts into sanctioned workflows, analytical utilities into versioned assets, and experienced architects’ judgment into institutional knowledge that remains available to younger staff.
Yule concludes that the greatest AI advantage may not belong to firms deploying the most agents. Instead, value will come from creating reusable, auditable systems that preserve organizational knowledge. The real opportunity lies in turning AI-assisted work into trusted tools and institutional memory.