CORTEXA
← Browse
arxivcs.AIcs.LGcs.MA2026-06-25

Semantic Early-Stopping for Iterative LLM Agent Loops

Sahil Shrivastava

Multi-agent large language model (LLM) loops, for example a Writer that drafts and a Critic that revises, are almost always terminated by a fixed iteration cap (max_iterations). This is a syntactic kill-switch: it is blind to whether the answer is still improving, so it over-spends tokens on easy inputs and truncates hard ones. We study semantic early-stopping: the loop halts when consecutive draft embeddings stop changing in meaning (cosine distance with a patience window) and the answer's measured quality stops improving. Our work makes three contributions. First, an honest theoretical footing: we prove deterministic termination and well-definedness and machine-check these claims, while treating the convergence of the distance sequence as an empirically tested conjecture rather than a (previously over-claimed) Banach contraction. Second, a judge-efficient evaluation protocol: we generate each question's full trajectory once, replay every stopping policy over the identical drafts, and cache every LLM-judge call, yielding a strictly paired efficiency-versus-quality comparison at low cost; we further separate operational tokens (charged to a policy) from evaluation tokens (a measurement instrument). Third, an empirical study on multi-hop retrieval-augmented question answering (HotpotQA). On the 60-question test split, a judge-free semantic stopper reduces operational tokens by 38% relative to max_iterations at parity quality (Delta-IS = -0.004, p = 0.81), whereas the full quality-gated variant is counter-productive because its per-round judging dominates cost. An oracle that selects the best round attains +0.115 Information Score over every practical policy (p ~ 4e-11), reframing the problem from "when to stop" (easy) to "which round is best" (open).

View free PDFSource page

Related papers

arxivcs.AIcs.CLcs.LGcs.MA2026-07-02

What LLM Agents Say When No One Is Watching: Social Structure and Latent Objective Emergence in Multi-Agent Debates

Arman Ghaffarizadeh, Danyal Mohaddes, Aliakbar Izadkhah, Shahriar Noroozizadeh

LLM agents will increasingly act in socially structured settings where role, audience, and relational context can shape what is advantageous or costly to say. We study whether such social structure, without any explicit objective in the prompt, changes what an agent expresses pub…

View free PDFSource page
arxivcs.CRcs.AIcs.IRcs.LGcs.MA2026-07-08

Who Broke the System? Failure Localization in LLM-Based Multi-Agent Systems

Yufei Xia, Anjun Gao, Yueyang Quan, Zhuqing Liu, Minghong Fang

Large language model (LLM) based multi-agent systems enable complex problem solving through coordinated reasoning and action, but their distributed structure also introduces new challenges in diagnosing system-level failures. When an execution fails, identifying which agent is re…

View free PDFSource page
arxivcs.CLcs.AIcs.LGcs.MA2026-06-28

Hybrid Retriever Evolution for Multimodal Document Reasoning Agents

Bohan Yao, Shruthan Radhakrishna, Vikas Yadav

Different retrievers, including lexical, semantic, and multimodal approaches, provide highly complementary strengths for multimodal document understanding, yet most systems combine them through fixed pipelines that cannot adapt to the demands of individual reasoning steps. In thi…

View free PDFSource page
arxivcs.AIcs.CLcs.LGcs.MA2026-07-10

AppWorld-UL: Benchmarking Diverse Agent-User Interactions for Tool-Use

Junzhi Chen, Harsh Trivedi, Jane Pan, Michael JQ Zhang, Tejas Srinivasan, Niranjan Balasubramanian, et al.

Tool-use agents that address day-to-day digital tasks such as ordering groceries must not only operate applications, but also interact with the user, e.g., to ask clarification questions, prompt for confirmation, and inform the user when the instruction is infeasible. However, cu…

View free PDFSource page
arxivcs.LGcs.AIcs.LOcs.MAcs.PL2026-07-06

PatchOptic for Shared-State LLM Workflows with Projected Views and Verified Structured Updates

Zhaoyu Bai, Jiaqi Cai

Agentic workflows often operate over shared, structured state. Because LLM context windows are limited, each model invocation is typically shown only the state fragment needed for the current workflow step, a pattern commonly known as progressive disclosure. Modern systems constr…

View free PDFSource page
arxivcs.AIcs.ETcs.LGcs.MAecon.GN2026-07-09

SolarChain-Eval: A Physics-Constrained Benchmark for Trustworthy Economic Agents in Decentralized Energy Markets

Shilin Ou, Yifan Xu, Luyao Zhang

As agentic AI systems are increasingly applied to cyber-physical environments, their evaluation requires assessment of both task performance and trustworthiness. In decentralized energy markets, autonomous agents may improve market utility, but may also exploit invalid physical d…

View free PDFSource page