Agentic systems promise autonomy and reasoning, but most recent architectures still treat them as collections of prompts, tools, and loosely defined agents. The result is often chaotic: unclear boundaries, fragile integrations, and reasoning that becomes difficult to govern or even understand.

In traditional software architecture, we faced a similar challenge when systems became too complex to manage. The solution was Domain-Driven Design, which introduced boundaries, language, and models to structure business logic.

But agentic systems introduce a new problem: reasoning itself becomes part of the architecture.
So how should we structure it?

This talk explores what happens when we apply architectural thinking to reasoning systems themselves. By looking at how knowledge and reasoning are structured in software, science, and even human cognition, we’ll examine how agentic systems can move beyond ad-hoc tool orchestration toward architectures with clear boundaries, responsibilities, and integration patterns.

If agentic systems are to become reliable architectural components rather than experimental prototypes, we may need to structure reasoning with the same discipline we use to structure business domains.