Skip to content
aiqbench takeAI Research

aiqbench analysis

This research establishes the Execution Information Requirement (EIR) as a critical metric for evaluating how long-running AI agents preserve and recover essential state information across task steps. By introducing frameworks like LACUNA and VESTIGE, the study demonstrates that mere storage capacity is insufficient; retention policies and error propagation mechanisms fundamentally dictate agent success rates in complex workflows. These findings provide a rigorous foundation for designing robust stateful agents capable of maintaining coherence over extended execution distances.
Sources coveringarXiv, Effective Altruism Forum, mignews.com, LessWrong
What this isA model's reading of the article and related coverage, written for aiqbench. Not the publisher's words.

Dude, Where's My State? Execution Information Requirements for Stateful Agents

arXiv2 min read

arXiv:2609.32687v1 Announce Type: new Abstract: Long-running agents must preserve information that later steps depend on. We introduce the Execution Information Requirement (EIR), a lower bound on the information that must remain accessible for correct completion under specified task and access conditions. We develop LACUNA, a framework that generates tasks with known dependencies and varies information demand, retention, and recovery separately from the difficulty of individual operations. Across four models, restoring a missing result raises accuracy on affected recall steps to 100%, compared with 0% for equal-length irrelevant information. Sufficient storage alone does not ensure success: retention policies can discard required results, errors can propagate through later computations, and agents can stop before recovery is complete. We also introduce VESTIGE, which uses agent execution traces to construct semantic graphs and measure information demand for real tasks. Across 72,562 software-agent trajectories, VESTIGE reveals a steeper distance-related decline in solution-relevant rereading for failed runs (RR 0.951 per distance doubling), while adjusted peak demand alone is not associated with failure. Together, these contributions support evaluating whether agents preserve and recover the information their tasks require.

Title:Dude, Where's My State? Execution Information Requirements for Stateful Agents

Abstract:Long-running agents must preserve information that later steps depend on. We introduce the Execution Information Requirement (EIR), a lower bound on the information that must remain accessible for correct completion under specified task and access conditions. We develop LACUNA, a framework that generates tasks with known dependencies and varies information demand, retention, and recovery separately from the difficulty of individual operations. Across four models, restoring a missing result raises accuracy on affected recall steps to 100%, compared with 0% for equal-length irrelevant information. Sufficient storage alone does not ensure success: retention policies can discard required results, errors can propagate through later computations, and agents can stop before recovery is complete. We also introduce VESTIGE, which uses agent execution traces to construct semantic graphs and measure information demand for real tasks. Across 72,562 software-agent trajectories, VESTIGE reveals a steeper distance-related decline in solution-relevant rereading for failed runs (RR 0.951 per distance doubling), while adjusted peak demand alone is not associated with failure. Together, these contributions support evaluating whether agents preserve and recover the information their tasks require.

35 pages, 12 figures, 15 tables. Supplementary material included

Artificial Intelligence (cs.AI)

arXiv:2609.32687 [cs.AI]

(or arXiv:2609.32687v1 [cs.AI] for this version)

https://doi.org/10.48550/arXiv.2609.32687 Focus to learn more arXiv-issued DOI via DataCite (pending registration)

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Full text under the publisher's open licence · 289 words

Related coverage