AI RESEARCH
A Trace-Based Assurance Framework for Agentic AI Orchestration: Contracts, Testing, and Governance
arXiv CS.AI
•
ArXi:2603.18096v1 Announce Type: cross In Agentic AI, Large Language Models (LLMs) are increasingly used in the orchestration layer to coordinate multiple agents and to interact with external services, retrieval components, and shared memory. In this setting, failures are not limited to incorrect final outputs. They also arise from long-horizon interaction, stochastic decisions, and external side effects (such as API calls, database writes, and message sends