Beyond Autonomy: A Dynamic Tiered AgentRunner Framework for Governable and Resilient Enterprise AI Execution
2026-05-11 • Artificial Intelligence
Artificial IntelligenceSoftware Engineering
AI summaryⓘ
The authors identify that current large language model agents focus on working independently but lack proper controls for safe use in businesses. They introduce the Dynamic Tiered AgentRunner, which changes how tasks are handled based on how risky they are, ensuring safer and more efficient outcomes. Their system uses separate agents for suggesting, reviewing, executing, and checking work, and includes a way to quickly detect and fix problems. This approach aims to balance safety, efficiency, and reliability in automated language model tasks.
large language modelagent frameworkrisk managementtiered executionseparation of powersverificationmulti-tenant SaaSresource allocationfault toleranceautomation governance
Authors
Kai Pan, Rong Hou
Abstract
Current large language model agent frameworks prioritize autonomy but lack the governability mechanisms required for enterprise deployment. High-risk write operations proceed without independent review, complex tasks lack acceptance verification, and computational resources are allocated uniformly regardless of risk level. We propose the Dynamic Tiered AgentRunner, a controlled execution protocol distilled from a production-grade multi-tenant SaaS platform. The framework introduces three core mechanisms: (1) Risk-Adaptive Tiering that dynamically allocates computational resources and review intensity based on task risk profiles, achieving Pareto-optimal trade-offs between safety and efficiency; (2) Separation of Powers architecture where proposal, review, execution, and verification are performed by independent agents with physically isolated boundaries; and (3) Resilience-by-Design through a Verifier-Recovery closed loop that treats failure as a first-class system state. We formalize the tier selectio