TimeWarp: Evaluating Web Agents by Revisiting the Past
This work addresses the problem of web agents' fragility to UI changes, which is crucial for developers and users relying on robust automation.
This paper introduces TimeWarp, a benchmark that simulates the evolving web to evaluate web agents' robustness to UI changes. Experiments show that existing agents are vulnerable to these changes, with Qwen-3 4B dropping from 20.4% to 0% and Llama-3.1 8B from 0% to 0% on new versions. The authors propose TimeTraj, a plan distillation algorithm, which improves Qwen-3 4B performance from 20.4% to 37.7% and Llama-3.1 8B from 0% to 27.0% on the benchmark.
The improvement of web agents on current benchmarks raises the question: Do today's agents perform just as well when the web changes? We introduce TimeWarp, a benchmark that emulates the evolving web using containerized environments that vary in UI, design, and layout. TimeWarp consists of three web environments, each with six UI versions spanning different eras of the internet, paired with a set of complex, realistic tasks requiring different forms of web navigation. Our experiments reveal web agents' vulnerability to changes and the limitations of behavior cloning (BC) on single-version trajectories. To address this, we propose TimeTraj, a simple yet effective algorithm that uses plan distillation to collect trajectories across multiple versions. By training agents on teacher rollouts using our BC-variant, we achieve substantial performance gains: $20.4\%\rightarrow37.7\%$ for Qwen-3 4B and $0\%\rightarrow27.0\%$ for Llama-3.1 8B models. We hope our work helps researchers study generalization across web designs and unlock a new paradigm for collecting plans rather than trajectories, thereby improving the robustness of web agents.