A new arXiv preprint introduces a method for making web agents more token-efficient by modeling how web applications behave. The approach aims to reduce the cost of processing tokens, which the authors note is rising quickly alongside investment in agentic infrastructure. The paper is titled "Token Efficient Task Execution via Application Behavior Modeling for Web Agents."
Researchers present TimeWarp, a benchmark designed to reflect how websites change over time, addressing the concern that agents may not generalize from today's web to tomorrow's. It comprises three web environments that emulate an evolving web so agent performance can be measured under such shifts. The work appears as a replacement submission on arXiv.
A new arXiv paper introduces AutoTailor, a meta-agentic framework that automatically selects and adapts reusable capabilities for web agents. The work targets a common problem: tool collections discovered automatically tend to be large, redundant, and not well matched to what users actually need. By aligning capability choice with user demand, the authors aim to cut the overhead of low-level browser interaction.
A new arXiv paper investigates what happens to language agents when the routines they have memorized no longer fit their environment. Combining a retrospective, human-assisted analysis with controlled web-task experiments, the authors test when stored procedures can still be successfully reused and when they interfere with one another. The work addresses a core assumption behind procedural memory in autonomous agents.