Composing Verifiable Conceptual Models via Building Blocks: Towards Design-Time Verification of Agentic AI Workflows

The article presents a design-time verification approach for agentic AI workflows, focusing on the orchestration of multiple LLM-based agents. It introduces a set of twelve structural rules to assess the compatibility of reusable building blocks within these workflows, implemented in a software prototype. Evaluation on 48 flawed workflows and 168 structurally altered variants demonstrated the verifier's effectiveness in detecting design violations, highlighting its potential for enhancing the safety and reliability of AI systems during the design phase.