Agentic Test-Time Scaling for WebAgents

2026-02-12

AI Summary

This work addresses the performance degradation of multi-step agents in long-horizon tasks due to error accumulation and the diminishing returns of uniformly increasing inference computation. To this end, the authors propose CATTS, a novel approach that introduces, for the first time, a dynamic computation allocation mechanism based on voting uncertainty into multi-step web agents. By leveraging an LLM arbitrator and a voting aggregation strategy—augmented with uncertainty metrics such as entropy and top-1/top-2 logit gaps—the method dynamically allocates additional computational resources when decision disagreement is high. Evaluated on WebArena-Lite and GoBrowse, CATTS achieves up to a 9.1% higher success rate over ReAct while reducing token consumption by up to 2.3× compared to uniform scaling, thereby enabling efficient and interpretable test-time scaling.

Technology Category

• Multiagent Systems: Adversarial Agents
• Reasoning under Uncertainty: Sequential Decision Making
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Test-time scaling has become a standard way to improve performance and boost reliability of neural network models. However, its behavior on agentic, multi-step tasks remains less well-understood: small per-step errors can compound over long horizons; and we find that naive policies that uniformly increase sampling show diminishing returns. In this work, we present CATTS, a simple technique for dynamically allocating compute for multi-step agents. We first conduct an empirical study of inference-time scaling for web agents. We find that uniformly increasing per-step compute quickly saturates in long-horizon environments. We then investigate stronger aggregation strategies, including an LLM-based Arbiter that can outperform naive voting, but that can overrule high-consensus decisions. We show that uncertainty statistics derived from the agent's own vote distribution (entropy and top-1/top-2 margin) correlate with downstream success and provide a practical signal for dynamic compute allocation. Based on these findings, we introduce Confidence-Aware Test-Time Scaling (CATTS), which uses vote-derived uncertainty to allocate compute only when decisions are genuinely contentious. CATTS improves performance on WebArena-Lite and GoBrowse by up to 9.1% over React while using up to 2.3x fewer tokens than uniform scaling, providing both efficiency gains and an interpretable decision rule.

Problem

• test-time scaling
• web agents
• multi-step tasks
• compute allocation
• decision uncertainty

Innovation

• Test-Time Scaling
• Agentic Reasoning
• Dynamic Compute Allocation
• Uncertainty Estimation
• Web Agents