$\tau^2$-Bench: Evaluating Conversational Agents in a Dual-Control Environment

TL;DR

This paper introduces $ au^2$-bench, a new benchmark for evaluating conversational agents in a dual-control environment where both agent and user actively modify a shared world, addressing limitations of existing single-control benchmarks.

Contribution

The paper presents a novel dual-control domain modeled as a Dec-POMDP, a compositional task generator, a reliable user simulator, and detailed performance analysis methods for conversational agents.

Findings

Significant performance drops when shifting from no-user to dual-control scenarios.

The benchmark effectively tests agent coordination and communication in shared environments.

The environment highlights challenges in guiding users and reasoning under dual-control conditions.

Abstract

Existing benchmarks for conversational AI agents simulate single-control environments, where only the AI agent can use tools to interact with the world, while the user remains a passive information provider. This differs from real-world scenarios like technical support, where users need to actively participate in modifying the state of the (shared) world. In order to address this gap, we introduce $\tau^2$-bench, with four key contributions: 1) A novel Telecom dual-control domain modeled as a Dec-POMDP, where both agent and user make use of tools to act in a shared, dynamic environment that tests both agent coordination and communication, 2) A compositional task generator that programmatically creates diverse, verifiable tasks from atomic components, ensuring domain coverage and controlled complexity, 3) A reliable user simulator tightly coupled with the environment, whose behavior is constrained by tools and observable states, improving simulation fidelity, 4) Fine-grained analysis of agent performance through multiple ablations including separating errors arising from reasoning vs communication/coordination. In particular, our experiments show significant performance drops when agents shift from no-user to dual-control, highlighting the challenges of guiding users. Overall, $\tau^2$-bench provides a controlled testbed for agents that must both reason effectively and guide user actions.