Dynamic Tiered AgentRunner: A Framework for Governable Enterprise AI

A new framework called the Dynamic Tiered AgentRunner proposes a controlled execution protocol for large language model agents in enterprise settings, addressing governance gaps in current autonomous systems. The framework introduces three core mechanisms: Risk-Adaptive Tiering for dynamic resource allocation based on task risk, a Separation of Powers architecture with independent agents for proposal, review, execution, and verification, and a Resilience-by-Design Verifier-Recovery closed loop. Developed from a production-grade multi-tenant SaaS platform, it aims to achieve Pareto-optimal trade-offs between safety and efficiency. The paper is available on arXiv under identifier 2605.10223.