AI agents strive for autonomy. Enterprise systems require control.
Purely LLM-driven architectures reach limits in quality, cost, and maintainability. An alternative approach is hybrid systems: deterministic domain models and business rules combined with probabilistic LLM components.
The talk shows how strong domain modeling serves as a guardrail: inputs and outputs are structured, validated, and deliberately controlled instead of letting agents figure everything out on their own.
The added value and limitations of LLMs are discussed, along with sensible responsibility allocation and integration patterns between agents and classical models.
The goal of the talk is to present practical architectural approaches that make it possible to integrate AI into existing system landscapes in a controlled, maintainable, and economical way.