Composing Agents to Minimize Worst-case Risk

2025-06-05

AI Summary

This work addresses robustness challenges in deploying multi-agent AI systems by proposing a novel paradigm for joint optimization of task success probability and tail risk (Value-at-Risk, VaR) over agent graphs—directed acyclic workflow structures. We formalize worst-case risk minimization as a VaR-path optimization problem on agent graphs, the first such formulation. Our method integrates agent-graph modeling, tail-risk quantification, probabilistic bound approximation via union bounds, and reinforcement learning–based agent composition. We design a near-optimal algorithm combining dynamic programming with union-bound analysis, and prove its asymptotic optimality for general loss functions. Evaluated on a video-game-style control benchmark, our approach significantly enhances risk controllability, accurately approximates VaR, and consistently identifies robust agent compositions that achieve both high success rates and low tail risk.

Technology Category

• Multiagent Systems: Adversarial Agents
• Reasoning under Uncertainty: Stochastic Optimization
• Intelligent Robots: Learning & Optimization for ROB

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Agentic search

Abstract

From software development to robot control, modern agentic systems decompose complex objectives into a sequence of subtasks and choose a set of specialized AI agents to complete them. We formalize an agentic workflow as a directed acyclic graph, called an agent graph, where edges represent AI agents and paths correspond to feasible compositions of agents. When deploying these systems in the real world, we need to choose compositions of agents that not only maximize the task success, but also minimize risk where the risk captures requirements like safety, fairness, and privacy. This additionally requires carefully analyzing the low-probability (tail) behaviors of compositions of agents. In this work, we consider worst-case risk minimization over the set of feasible agent compositions. We define worst-case risk as the tail quantile -- also known as value-at-risk -- of the loss distribution of the agent composition where the loss quantifies the risk associated with agent behaviors. We introduce an efficient algorithm that traverses the agent graph and finds a near-optimal composition of agents by approximating the value-at-risk via a union bound and dynamic programming. Furthermore, we prove that the approximation is near-optimal asymptotically for a broad class of practical loss functions. To evaluate our framework, we consider a suite of video game-like control benchmarks that require composing several agents trained with reinforcement learning and demonstrate our algorithm's effectiveness in approximating the value-at-risk and identifying the optimal agent composition.

Problem

• Formalizing agentic workflows as graphs to optimize task success and risk
• Minimizing worst-case risk via tail quantile analysis of agent compositions
• Developing efficient algorithms for near-optimal agent composition selection
• Minimize worst-case risk in agentic workflows
• Optimize agent compositions for safety and performance
• Analyze tail behaviors of AI agent combinations

Innovation

• Formalizes agentic workflows as directed acyclic graphs
• Minimizes worst-case risk via tail quantile approximation
• Uses dynamic programming for near-optimal agent composition
• Agent graph models workflows as DAGs
• Minimizes worst-case risk via tail quantiles
• Efficient dynamic programming for near-optimal compositions