Willful Disobedience: Automatically Detecting Failures in Agentic Traces

TL;DR

AgentPex is an AI tool that systematically evaluates agentic traces for procedural failures and specification violations, improving validation beyond outcome-only benchmarks.

Contribution

This paper introduces AgentPex, a novel system that extracts behavioral rules from prompts to automatically detect violations in agentic traces.

Findings

AgentPex distinguishes agent behavior across different models.

It surfaces specification violations not captured by outcome-only scoring.

Provides fine-grained analysis by domain and metric.

Abstract

AI agents are increasingly embedded in real software systems, where they execute multi-step workflows through multi-turn dialogue, tool invocations, and intermediate decisions. These long execution histories, called agentic traces, make validation difficult. Outcome-only benchmarks can miss critical procedural failures, such as incorrect workflow routing, unsafe tool usage, or violations of prompt-specified rules. This paper presents AgentPex, an AI-powered tool designed to systematically evaluate agentic traces. AgentPex extracts behavioral rules from agent prompts and system instructions, then uses these specifications to automatically evaluate traces for compliance. We evaluate AgentPex on 424 traces from ${\tau}^2$-bench across models in telecom, retail, and airline customer service. Our results show that AgentPex distinguishes agent behavior across models and surfaces specification violations that are not captured by outcome-only scoring. It also provides fine-grained analysis by domain and metric, enabling developers to understand agent strengths and weaknesses at scale. The source code of AgentPex is available at https://github.com/microsoft/agentpex.