Beyond Autonomy: A Dynamic Tiered AgentRunner Framework for Governable and Resilient Enterprise AI Execution
TLDR
Dynamic Tiered AgentRunner enhances enterprise AI with governability, resilience, and risk-adaptive execution, moving beyond pure autonomy.
Key contributions
- Introduces Risk-Adaptive Tiering for dynamic resource and review allocation based on task risk profiles.
- Employs a Separation of Powers architecture with independent agents for proposal, review, execution, and verification.
- Achieves Resilience-by-Design via a Verifier-Recovery closed loop, treating system failures as first-class states.
Why it matters
Current LLM agent frameworks lack the governability needed for enterprise deployment, leading to issues like unreviewed high-risk operations. This paper introduces a framework that provides essential mechanisms for safety, efficiency, and resilience. It enables controlled, production-grade AI execution by addressing these critical gaps.
Original 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
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.