Pith. sign in

REVIEW 4 major objections 5 minor 142 references

The Horizon Gap: Planning, Memory, Execution, Training, and Evaluation for Long-Horizon LLM Agents

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper claims the field's scattered long-horizon agent research is one coordinated response to signal sparsity: as horizons grow, outcome-only signals become uninformative, and training and evaluation both manufacture denser step-level…

desk verdict A serious, unusually transparent survey that names a real gap and builds a useful taxonomy; the corpus behind it needs to be public and independently labeled before the empirical claims carry full weight. read the letter →

arxiv 2608.06663 v1 pith:OJ24EH3U submitted 2026-08-07 cs.CL

Mingguang Chen , Licheng Wang , Bo Qu This is my paper · ORCID
classification cs.CL
keywords horizongaplong-horizonagentslong-contextlong-termmemoryprocessrewardmodelstrajectory-levelevaluationharnessvsmodelcorrelatedmeasurementbias
topics The Measurement Problem
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper identifies a "horizon gap": large language models solve single-step problems well but fail at tasks spanning hours, losing track of earlier decisions, declaring unfinished work done, and drifting from goals. It surveys 1,547 papers from 2024 to 2026 and argues that the field's many strands are one coordinated response to a single pressure: as horizons grow, outcome-only signals (one final reward, one pass/fail check) become uninformative, so researchers manufacture denser step-level signals for training and evaluation alike. The paper disambiguates long-horizon (a task property), long-context (a model property), and long-term memory (a system property), then organizes the corpus into six lifecycle categories crossed with where the horizon is carried. A sympathetic reader would care because the survey turns scattered agent research into a testable structural claim and names two concrete measurement problems that block progress: separating model capability from harness capability, and avoiding correlated bias between training and evaluation signals.

What carries the argument

The organizing device is a two-axis taxonomy: six lifecycle categories (planning, memory, execution, training, evaluation, foundations) crossed with where the horizon is carried (within-context, within-task-beyond-context, cross-task-persistent). The load-bearing analytical identity is the structural analogy between training and evaluation: both need a reliable signal of partial-trajectory progress, and both manufacture it at the step level, which creates the risk of correlated measurement bias. The paper's evidence base is a systematically harvested corpus with a disclosed bleed filter; the ratio external:context memory (294:103) and the category growth timeline are the concrete observations carrying the synthesis.

What would settle it

Re-run the full 1,547-paper classification with a second independent annotator using the released scripts and compare per-category ratios; if, say, the external-to-context memory ratio of 294:103 or the Figure 6 growth timeline changes materially, the survey's structural pattern is an artifact of labeling. Alternatively, the harness-versus-model claim would be settled by a controlled experiment holding the model fixed across harnesses and holding the harness fixed across models and measuring achievable task horizon at a fixed reliability: if harness changes move the horizon little, the paper's central attribution problem dissolves.

Watch

Extended reading notes

Core claim

The central claim is that outcome-only supervision and outcome-only evaluation stop working as task horizon grows, and the field's response is the same everywhere: replace the single terminal signal with denser, step-level signal. In training this appears as process reward models and credit assignment; in evaluation as trajectory-level diagnostics and benchmark-purification work; in memory and execution as traceable, recoverable trajectories. The paper further claims the execution trajectory is becoming the shared unit of analysis across all six categories, and that this convergence makes trajectory logging a precondition for answering the field's two open measurement problems.

Load-bearing premise

The corpus is representative and its labels are trustworthy; classification rested on a single annotator's rules, with residual misclassification estimated from only 85 manually re-checked papers, so systematic labeling errors would invalidate the category ratios and growth timeline on which the synthesis rests.

Editorial extensions

If this is right

  • If the thesis is right, improving long-horizon reliability means investing in step-level signal design for both training and evaluation, not just scaling models or contexts.
  • The harness-versus-model question must be answered before long-horizon capability can be rationally pursued; a controlled same-model/different-harness comparison is the implied decisive experiment.
  • The convergence on trajectory as the unit of analysis implies trajectory logging becomes infrastructure, not an implementation detail.
  • Benchmark critiques (leakage, weak tests, coherence collapse) are not side literature; they are first-class evidence that outcome scores overstate capability.
  • Training algorithms described as outcome-based, such as GRPO, may already perform implicit process-level credit assignment, so the outcome/process distinction is not architectural.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If correlated measurement bias is real, published long-horizon progress may be systematically overstated, and a decisive test would compare process signals built from disjoint assumptions.
  • The taxonomy predicts research attention follows visible failure; a testable extension is whether foundations-paper counts lag capability demonstrations by a roughly constant delay.
  • Because the corpus was labeled by a single annotator with rule-based defaults, all category ratios should be treated as hypotheses; an independent re-labeling of the same corpus would be a cheap check.
  • If trajectory retention becomes the norm, privacy and auditability of stored traces become load-bearing, connecting the memory-security thread to the oversight thread.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. This survey defines 'the horizon gap' as the distance between single-step model capability and reliable completion of tasks spanning many steps, and maps the 2024-2026 arXiv literature responding to it. The corpus (1,547 papers) is built via an eight-thread seed harvest with a disclosed two-stage bleed filter (26.8% excluded), supplemented by 128 targeted papers for under-covered foundations/safety work. The paper disambiguates long-horizon (task property), long-context (model property), and long-term memory (system property); organizes the corpus into six lifecycle categories (planning, memory, execution, training, evaluation, foundations) crossed with a horizon-locus axis (within-context, within-task-beyond-context, cross-task-persistent); and advances a cross-cutting thesis that as horizon grows, outcome-only signals become uninformative and the field responds by manufacturing denser step-level signals, visible in process reward models and credit assignment (training) and in trajectory-level diagnostics and benchmark-critique work (evaluation). It closes by naming three open measurement problems: model-versus-harness attribution, correlated measurement bias between training and evaluation signals, and whether long-horizon reliability admits a general predictive theory.

Significance. Should the synthesis hold, the central contribution is a reframing: six sub-literatures of long-horizon agent research read as one coordinated response to signal sparsity, together with a clean tripartite disambiguation (long-horizon vs long-context vs long-term memory) that is itself genuinely useful, and two open measurement problems (model-versus-harness attribution; correlated measurement bias in process-level signals) concrete enough to guide future experiments. The paper is internally consistent: I verified the corpus arithmetic (1,939 unique raw hits; 520 dropped over two stages = 26.8%; 1,419 seed + 128 supplement = 1,547; Table 1's categories and subcategories sum exactly to the corpus total, and the supplement distribution across categories sums to 128). The methodology is unusually transparent for the genre — disclosed bleed rates with quantified per-stage drops, an explicit rule-plus-override classification procedure, hedged 'hypothesis, not finding' language for corpus-shape observations, and one genuinely falsifiable prediction (the verifiability-gradient delay).

major comments (4)
  1. [§2.2, Table 1, Figure 6] The quantitative texture of the survey — per-category counts, the external:context (294:103), orchestration:recovery (338:245), and rl:supervision (130:37) ratios used in the section Assessments, and the Figure 6 growth timeline — rests on a single annotator's pipeline validated by a manual re-check of only 85 papers (30 kept, 55 excluded). The reported residual error ('on the order of one in twenty') is a point estimate with no confidence interval, no per-category breakdown, and no second annotator; because the re-check was performed by the same annotator against their own criterion, it establishes rule-consistency rather than construct validity against an independent expert. The headline ratios are robust to uniform ±5% noise, but systematic directional bias (e.g., a broadened 'external' tag absorbing in-context work, or an over-broad 'orchestration' tag) is exactly the failure mode the validation cannot detect at this scale, and it would change the conclusions the Assessments draw. The skeptic's worry that misclassification could manufacture the reported pattern is, on reading the paper, only partially warranted: the paper's own hedging and the independently cited literature (the PRM800K result, the SWE-bench+ critique, the trajectory-level diagnostics of §7) would survive substantial label noise. But the per-category ratios and the §9 corpus-shape hypotheses are as fragile as the skeptic says. Because the corpus is the paper's primary empirical artifact, I ask that the corpus and scripts be released at submission rather than at camera-ready, and that the validation be extended to per-category residual-error estimates with intervals and, ideally, an independent second-annotator pass on a stratified sample.
  2. [Abstract, §9, §10] The abstract's central claim — 'Across all categories, we find the same pattern: outcome-only signals grow uninformative as horizons lengthen' — is stated as a universal, and §9 repeats the universal form ('Across every category, the object being stored, scored, attributed over, audited, and debugged is shifting from the outcome to the execution trace'). The body supports the signal-density version of the claim strongly only for training (§6) and evaluation (§7). The §3 assessment describes a commitment-versus-robustness trade-off, the §4 assessment a persistence-versus-fidelity trade-off, and the §5 assessment a harness-versus-model attribution pattern; these are structural responses to horizon growth, but they are not the same claim about outcome-only signals growing uninformative, and §10 itself narrows the outcome-to-process shift to training and evaluation. To make the headline claim match the evidence, either restrict the abstract/conclusion formulation to the two well-supported instantiations plus a clearly-labeled broader 'densification of structure' reading, or supply per-category evidence that signal density specifically is the cross-cutting response.
  3. [Table 1, §9] §9's second interpretive hypothesis (the 'verifiability-gradient' account) argues from Table 1's 2026-share column — foundations at 77%, memory at 72% — but that column includes the 128 supplement papers, which §2.2 declares 'recency-biased by construction' and which are excluded from Figure 6 for precisely that reason. For foundations, 93 of 103 papers are supplement papers harvested with dedicated 2024+ queries, so the 77% figure is substantially an artifact of the supplement's construction; using it as evidence for a substantive claim about the field's recent priorities is inconsistent with the paper's own exclusion of the supplement from the growth analysis. The 2026-share column should be recomputed for seed-only papers, or the hypothesis should be argued from seed-only shares.
  4. [§2.2, §2.3] The taxonomy and the counts are entangled: the exploratory k-means clustering over the seed pool supplied the 'cluster-based category defaults' that the classification then used, so the category boundaries — and hence Table 1's counts and Figures 2 and 6 — were fit to the same data they are then used to summarize. The paper discloses this in §2.2, which is to its credit, but the section Assessments nevertheless treat several counts as findings (e.g., §4: 'the external-to-context size ratio... is itself informative'). Because the clustering defaults, keyword rules, and override list jointly determine the counts, the paper should (i) state at each point where a count is used as evidence that the count inherits the taxonomy's construction choices, and (ii) release the cluster assignments alongside the labels so the dependence can be audited.
minor comments (5)
  1. [Throughout] The text contains ligature artifacts ('efficient', 'efficiency', 'difficulty', 'sufficient' in §§2-8) and spacing artifacts ('Data A vailability', 'W ALL-E') that should be cleaned in the published version; they appear to be rendering artifacts rather than intentional formatting.
  2. [§2.2] The phrase 'rather than force the corpus down to a pre-registered size' is ambiguous: the earlier mention of 'our initial ~1,100–1,400 planning estimate' suggests an internal planning estimate, and readers should not infer a formal preregistration; please clarify the status of the estimate.
  3. [Figure 5] The caption states that the figure is 'Sourced entirely from primary papers' while the plotted bars mix 'reported or illustrative ranges'; the two classes of ranges should be distinguished visually and in the caption, and the cited artifact file (artifacts/benchmark_durations.csv) should be released with the corpus so the point estimates can be traced to their sources.
  4. [§§3-9] The prose synthesizes roughly 130 exemplar papers out of a 1,547-paper corpus, but the criterion for which papers become prose exemplars rather than corpus rows is not stated; one sentence describing the exemplar-selection principle (representativeness, influence, topical span, or recency) would help readers judge whether the narrative is anchored to a purposive sample.
  5. [§§6-8] Several load-bearing diagnostic findings cited in support of the synthesis are 2026 arXiv preprints without peer-review status (e.g., [76] on the accuracy-correction paradox, [92] on GRPO as an implicit process reward model, [108] on coherence collapse); the paper's language calibration is exemplary for its own claims, and the same discipline should be extended to first-mention status flags for cited preprints whose findings have not yet been independently replicated.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the survey contains no derived quantities that reduce to their inputs, corpus counts are descriptive outputs of a disclosed classification pipeline, and the sole self-citation is scoped, non-load-bearing positioning.

full rationale

This is a survey with no equations, fitted parameters, or quantitative predictions; the central synthesis (outcome-only signals grow uninformative as horizon lengthens, so the field manufactures denser step-level signals) is an inductive reading of the cited literature rather than a quantity derived from an input. The per-category counts (execution n=584, external memory 294 vs. context 103, orchestration 338 vs. recovery 245) are outputs of the authors' own taxonomy, but the categories are not defined in terms of those counts, and §2.2 discloses the single-annotator, rule-based pipeline, the bleed filters, and the residual misclassification estimate. Any concern about label reliability is a validity/correctness issue, not circularity: the taxonomy was not fitted to reproduce the pattern it is used to describe. The one self-citation, the companion recursive-self-improvement survey [15], is explicitly scoped in §2.4 and §6 as sharing harvest infrastructure but not scope, and it is not used to justify any conclusion. The paper even anticipates and explicitly rejects the closest circularity-adjacent worry in §9, stating that nothing observed shows a specific benchmark inheriting a specific training signal, and frames correlated measurement bias as an unproven structural hypothesis. Thus no load-bearing step reduces by construction to its own inputs.

Assumptions & free parameters 4 free parameters · 4 assumptions · 1 invented entities

This survey has no fitted physical parameters. The relevant hand-chosen design choices are the harvest caps, cluster count, and inclusion term lists, which shape the corpus composition that the paper's quantitative observations depend on.

free parameters (4)
  • Seed-thread result cap per query = 250
    Each of the eight arXiv API harvest threads was capped at 250 results, a disclosed depth limit that affects which papers enter the corpus; no saturation analysis is reported.
  • k-means cluster count = 18
    The exploratory clustering used k=18 with no stated model-selection criterion; the resulting clusters scaffold the taxonomy.
  • Harvest-time topic-signal term list
    Admission required at least one hand-chosen topic-signal term (e.g., long-horizon, multi-step, episod-, trajector-, planning, memory, credit assignment, reflection); the exact term list is described by examples only.
  • Second-stage gate exemption rule
    Agent-systems-specific clusters were exempted from the LLM/agent-signal gate by inspection after false-exclusion checks; the exemption criterion is qualitative, not a scored threshold.
assumptions (4)
  • domain assumption Cited findings are summarized accurately across 1,547 papers
    The synthesis treats each cited paper's stated result as reliable; no independent re-analysis of the cited experiments is performed.
  • domain assumption The arXiv harvest approximates the relevant literature
    The corpus excludes non-arXiv industrial engineering, non-English work, and blog or technical-report-only systems, which the paper acknowledges in §2.2.
  • domain assumption The eight seed threads span the main sub-directions
    The seed harvest presupposes this partition of the field; a different partition would change the corpus composition.
  • domain assumption The METR time-horizon trend is real but not settled
    The paper repeatedly flags the METR growth trend as measured and reported, not a law (§2.2, §7, §9), relying on the underlying NeurIPS 2025 measurement for part of its motivation.
invented entities (1)
  • The horizon gap
    purpose: Names the distance between single-step model capability and reliable multi-hour task completion; organizes the entire survey.
    A definitional framing construct, not a falsifiable entity; its measurable cousin, autonomy time, is imported from METR [1].

how reviews work

0 comments
Cite this review

Pith. "Pith review of The Horizon Gap: Planning, Memory, Execution, Training, and Evaluation for Long-Horizon LLM Agents." pith.science (2026). https://pith.science/paper/OJ24EH3U

@misc{pith2026260806663,
  author       = {Pith},
  title        = {Pith review of: The Horizon Gap: Planning, Memory, Execution, Training, and Evaluation for Long-Horizon LLM Agents},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OJ24EH3U}},
  note         = {Machine review of arXiv:2608.06663}
}
read the original abstract

Frontier language models solve reasoning problems in a single forward pass that would have been research contributions years ago, yet fail at multi-hour tasks: losing track of earlier decisions, declaring half-finished work done, or drifting from goals. We call this the horizon gap and survey 1,547 arXiv papers (2024-2026) collected via systematic seed harvest with a disclosed 26.8% bleed filter, extended by targeted supplementation. We disambiguate three routinely conflated properties: long-horizon (task property: required steps), long-context (model property: token capacity), and long-term memory (system property: persistence across steps/sessions). We organize the corpus into six categories tracking a long-horizon task's lifecycle -- planning, memory, execution, training, evaluation, and foundations/safety -- crossed with an axis capturing where horizons are carried (within-context, within-task-beyond-context, or cross-task-persistent). Across all categories, we find the same pattern: outcome-only signals grow uninformative as horizons lengthen, and the field's response -- whether process reward models, credit assignment, or trajectory-level diagnostics -- manufactures denser step-level signals. We treat critical and diagnostic literature as first-class threads throughout, arguing that segregating critique from method would routinely split single papers across chapters. We close by naming open measurement problems: decomposing model versus harness capability, managing correlated bias in process-level signals used for both training and evaluation, and whether long-horizon reliability admits general predictive theory.

Figures

Figures reproduced from arXiv: 2608.06663 by the authors.

Figure 1
Figure 1. Taxonomy grid – six categories (rows) crossed with the horizon-bearing-locus axis [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Semantic map of the corpus (TF-IDF + SVD + t-SNE projection, axes are arbitrary [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. The planning strategy spectrum – plan-then-execute, interleaved, search-based, and [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: The memory persistence-fidelity trade-off across three tiers – context, external store, and [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: Human time-to-complete one task instance across the small set of benchmarks that report [PITH_FULL_IMAGE:figures/full_fig_p022_5.png]
Figure 6
Figure 6. Figure 6: Growth timeline, seed corpus only (n=1,419). Top: quarterly paper counts per category, [PITH_FULL_IMAGE:figures/full_fig_p027_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

142 extracted references · 17 canonical work pages

  1. [1]

    Ziegler, Elizabeth Barnes, and Lawrence Chan

    Thomas Kwa, Ben West, Joel Becker, Amy Deng, Katharyn Garcia, Max Hasin, Sami Jawhar, Megan Kinniment, Nate Rush, Sydney Von Arx, Ryan Bloom, Thomas Broadley, Haoxing Du, Brian Goodrich, Nikola Jurkovic, Luke Harold Miles, Seraphina Nix, Tao Lin, Chris Painter, Neev Parikh, David Rein, Lucas Jun Koba Sato, Hjalmar Wijk, Daniel M. Ziegler, Elizabeth Barnes...

    arXiv 2025

  2. [2]

    Large language models cannot self-correct reasoning yet, 2024

    Jie Huang, Xinyun Chen, Swaroop Mishra, Huaixiu Steven Zheng, Adams Wei Yu, Xinying Song, and Denny Zhou. Large language models cannot self-correct reasoning yet, 2024. URL https://arxiv.org/abs/2310.01798. ICLR 2024

    arXiv 2024

  3. [3]

    Are ”solved issues” in SWE-bench really solved correctly? an empirical study, 2025

    You Wang, Michael Pradel, and Zhongxin Liu. Are ”solved issues” in SWE-bench really solved correctly? an empirical study, 2025. URL https://arxiv.org/abs/2503.15223

    arXiv 2025

  4. [4]

    The SWE-bench illusion: When state-of-the-art LLMs remember instead of reason, 2025

    Shanchao Liang, Spandan Garg, and Roshanak Zilouchian Moghaddam. The SWE-bench illusion: When state-of-the-art LLMs remember instead of reason, 2025. URL https://arxiv. org/abs/2506.12286

    arXiv 2025

  5. [5]

    Establishing best practices for building rigorous agentic benchmarks, 2025

    Yuxuan Zhu, Tengjun Jin, Yada Pruksachatkun, Andy Zhang, Shu Liu, Sasha Cui, Sayash Kapoor, Shayne Longpre, Kevin Meng, Rebecca Weiss, Fazl Barez, Rahul Gupta, Jwala Dhamala, Jacob Merizian, Mario Giulianelli, Harry Coppock, Cozmin Ududec, Jasjeet Sekhon, Jacob Steinhardt, Antony Kellermann, Sarah Schwettmann, Matei Zaharia, Ion Sto- ica, Percy Liang, and...

    arXiv 2025

  6. [6]

    Understanding the planning of LLM agents: A survey, 2024

    Xu Huang, Weiwen Liu, Xiaolong Chen, Xingmei Wang, Hao Wang, Defu Lian, Yasheng Wang, Ruiming Tang, and Enhong Chen. Understanding the planning of LLM agents: A survey, 2024. URL https://arxiv.org/abs/2402.02716

    arXiv 2024

  7. [7]

    LLMs as planning formalizers: A survey for leveraging large language models to construct automated planning models, 2025

    Marcus Tantakoun, Xiaodan Zhu, and Christian Muise. LLMs as planning formalizers: A survey for leveraging large language models to construct automated planning models, 2025. URL https://arxiv.org/abs/2503.18971

    arXiv 2025

  8. [8]

    A survey on the memory mechanism of large language model based agents, 2024

    Zeyu Zhang, Xiaohe Bo, Chen Ma, Rui Li, Xu Chen, Quanyu Dai, Jieming Zhu, Zhenhua Dong, and Ji-Rong Wen. A survey on the memory mechanism of large language model based agents, 2024. URL https://arxiv.org/abs/2404.13501

    arXiv 2024

Show all 142 references
  1. [9]

    Wei-Chieh Huang, Weizhi Zhang, Yueqing Liang, Yuanchen Bei, Yankai Chen, Tao Feng, Xinyu Pan, Zhen Tan, Yu Wang, Tianxin Wei, Shanglin Wu, Ruiyao Xu, Liangwei Yang, Rui Yang, Wooseong Yang, Chin-Yuan Yeh, Hanrong Zhang, Haozhen Zhang, Siqi Zhu, Henry Peng Zou, Wanjia Zhao, Son...

    2026

  2. [10]

    Chaoyun Zhang, Shilin He, Jiaxu Qian, Bowen Li, Liqun Li, Si Qin, Kang Yu, Minghua Ma, G. M. Liu, Qingwei Lin, Saravan Rajmohan, Dongmei Zhang, and Qi Zhang. Large language model-brained GUI agents: A survey, 2024. URL https://arxiv.org/abs/2411.18279

    2024 arXiv

  3. [11]

    Deanna D. Nguyen, Jian Chen, Yu Wang, Gang Wu, Namyong Park, Zhengmian Hu, Hanjia Lyu, Junda Wu, Ryan Aponte, Yu Xia, Xintong Li, Jing Shi, Hongjie Chen, Viet Dac Lai, Zhouhang Xie, Sungchul Kim, Ruiyi Zhang, Yu Tong, Mehrab Tanjim, Nesreen K. Ahmed, Puneet Mathur, Seunghyun Y...

    2024

  4. [12]

    A survey on (m)LLM-based GUI agents, 2025

    Fei Tang, Haolei Xu, Hang Zhang, Siqi Chen, Xingyu Wu, Yongliang Shen, Wenqi Zhang, Guiyang Hou, Zeqi Tan, Yuchen Yan, Kaitao Song, Jian Shao, Weiming Lü, Jun Xiao, and Yueting Zhuang. A survey on (m)LLM-based GUI agents, 2025. URL https://arxiv.org/ab s/2504.13865

    2025 arXiv

  5. [13]

    Guibin Zhang, Hejia Geng, Xiaohang Yu, Zhenfei Yin, Zaibin Zhang, Zelin Tan, Heng Zhou, Zhongzhi Li, Xiangyuan Xue, Yijiang Li, Yifan Zhou, Yang Chen, Chen Zhang, Yutao Fan, Zihu Wang, Songtao Huang, Francisco Piedrahita-Velez, Yue Liao, Hongru Wang, Mengyue Yang, Hong Ji, Jun...

    2025

  6. [14]

    When can LLMs actually correct their own mistakes? a critical survey of self-correction of LLMs, 2024

    Ryo Kamoi, Yusen Zhang, Nan Zhang, Jiawei Han, and Rui Zhang. When can LLMs actually correct their own mistakes? a critical survey of self-correction of LLMs, 2024. URL https: //arxiv.org/abs/2406.01297

    2024 arXiv

  7. [15]

    Recursive self-improvement in ai: From bounded self-refinement to autonomous research loops, 2026

    Mingguang Chen, Licheng Wang, and Bo Qu. Recursive self-improvement in ai: From bounded self-refinement to autonomous research loops, 2026. URL https://arxiv.org/ab s/2607.07663. Companion survey; DeepGrounding

    2026 arXiv

  8. [16]

    Sutton, Doina Precup, and Satinder Singh

    Richard S. Sutton, Doina Precup, and Satinder Singh. Between MDPs and semi-MDPs: A framework for temporal abstraction in reinforcement learning. Artificial Intelligence , 112 (1–2):181–211, 1999. doi: 10.1016/S0004-3702(99)00052-1. The Options Framework

    1999 doi

  9. [17]

    Peter Dayan and Geoffrey E. Hinton. Feudal reinforcement learning. In Advances in Neural Information Processing Systems 5 (NIPS 1992) , 1992. URL https://proceedings.neurips.cc /paper/1992/hash/d14220ee66aeec73c49038385428ec4c-Abstract.html

    1992

  10. [18]

    LLM+P: Empowering large language models with optimal planning proficiency, 2023

    Bo Liu, Yuqian Jiang, Xiaohan Zhang, Qiang Liu, Shiqi Zhang, Joydeep Biswas, and Peter Stone. LLM+P: Empowering large language models with optimal planning proficiency, 2023. URL https://arxiv.org/abs/2304.11477. arXiv preprint only

    2023 arXiv

  11. [19]

    Michael Ahn, Anthony Brohan, Noah Brown, Yevgen Chebotar, Omar Cortes, Byron David, Chelsea Finn, Chuyuan Fu, Keerthana Gopalakrishnan, Karol Hausman, Alex Herzog, Daniel 30 Ho, Jasmine Hsu, Julian Ibarz, Brian Ichter, Alex Irpan, Eric Jang, Rosario Jauregui Ruano, Kyle Jeffre...

    2022 arXiv

  12. [20]

    PoTable: Towards systematic thinking via plan-then-execute stage reasoning on tables, 2024

    Qingyang Mao, Qi Liu, Zhi Li, Mingyue Cheng, Zheng Zhang, and Rui Li. PoTable: Towards systematic thinking via plan-then-execute stage reasoning on tables, 2024. URL https://arxi v.org/abs/2412.04272

    2024 arXiv

  13. [21]

    Tree-of-code: A hybrid approach for robust complex task planning and execution, 2024

    Ziyi Ni, Yixuan Li, and Daxiang Dong. Tree-of-code: A hybrid approach for robust complex task planning and execution, 2024. URL https://arxiv.org/abs/2412.14212

    2024 arXiv

  14. [22]

    Furuta, Gopala K

    Lutfi Eren Erdogan, Nicholas Lee, Se Hoon Kim, Suhong Moon, H. Furuta, Gopala K. Anu- manchipalli, Kurt Keutzer, and Amir Gholami. Plan-and-act: Improving planning of agents for long-horizon tasks, 2025. URL https://arxiv.org/abs/2503.09572

    2025 arXiv

  15. [23]

    PIVOT: Bridging planning and execution in LLM agents via trajectory refinement, 2026

    Tuo Zhang, Alin-Ionut Popa, Yan Xu, Rui Song, and Dimitrios Dimitriadis. PIVOT: Bridging planning and execution in LLM agents via trajectory refinement, 2026. URL https://arxiv. org/abs/2605.11225

    2026 arXiv

  16. [24]

    Harnesses for inference-time alignment over execution trajectories, 2026

    Boyuan Wang, Bochao Li, Minghan Wang, Yuxin Tao, and Fang Kong. Harnesses for inference-time alignment over execution trajectories, 2026. URL https://arxiv.org/abs/ 2605.21516

    2026 arXiv

  17. [25]

    PRompt optimization in multi-step tasks (PROMST): Integrating human feedback and heuristic-based sampling, 2024

    Yongchao Chen, Jacob Arkin, Yilun Hao, Yang Zhang, Nicholas Roy, and Chuchu Fan. PRompt optimization in multi-step tasks (PROMST): Integrating human feedback and heuristic-based sampling, 2024. URL https://arxiv.org/abs/2402.08702

    2024 arXiv

  18. [26]

    ReAgent: Reversible multi-agent reasoning for knowledge-enhanced multi-hop QA, 2025

    Xinjie Zhao, Fan Gao, Xingyu Song, Yingjian Chen, Rui Yang, Yanran Fu, Yuyang Wang, Yusuke Iwasawa, Yutaka Matsuo, and Irene Li. ReAgent: Reversible multi-agent reasoning for knowledge-enhanced multi-hop QA, 2025. URL https://arxiv.org/abs/2503.06951

    2025 arXiv

  19. [27]

    Do agents need to plan step-by-step? rethinking planning horizon in data-centric tool calling, 2026

    Naoki Otani, Nikita Bhutani, Hannah Kim, Dan Zhang, and Estevam Hruschka. Do agents need to plan step-by-step? rethinking planning horizon in data-centric tool calling, 2026. URL https://arxiv.org/abs/2605.08477

    2026 arXiv

  20. [28]

    The cognitive bandwidth bottleneck: Shifting long-horizon agent from plan- ning with actions to planning with schemas, 2025

    Bo Xu, Tianshi Zheng, Zhaowei Wang, Hong Ting Tsang, Weiqi Wang, Tianqing Fang, and Yangqiu Song. The cognitive bandwidth bottleneck: Shifting long-horizon agent from plan- ning with actions to planning with schemas, 2025. URL https://arxiv.org/abs/2510.07091

    2025

  21. [29]

    Griffiths, Yuan Cao, and Karthik Narasimhan

    Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, and Karthik Narasimhan. Tree of thoughts: Deliberate problem solving with large language models. In Advances in Neural Information Processing Systems (NeurIPS) , 2023. URL https://arxiv.org/abs/2305.10601

    2023 arXiv

  22. [30]

    An, Yuwei Wan, Wenliang Chen, Binbin Lin, Yuqiang Li, Tong Xie, and Dongzhan Zhou

    Mengsong Wu, Yafei Wang, Yidong Ming, Y. An, Yuwei Wan, Wenliang Chen, Binbin Lin, Yuqiang Li, Tong Xie, and Dongzhan Zhou. CheMatAgent: Enhancing LLMs for chemistry 31 and materials science through tree-search based tool learning, 2025. URL https://arxiv.org/ abs/2506.07551

    2025 arXiv

  23. [31]

    SELT: Self- evaluation tree search for LLMs with task decomposition, 2025

    Mengsong Wu, Di Zhang, Yuqiang Li, Dongzhan Zhou, and Wenliang Chen. SELT: Self- evaluation tree search for LLMs with task decomposition, 2025. URL https://arxiv.org/abs/ 2506.07557

    2025 arXiv

  24. [32]

    W ALL-e: World alignment by rule learning improves world model-based LLM agents,

    Siyu Zhou, Tianyi Zhou, Yijun Yang, Guodong Long, Deheng Ye, Jing Jiang, and Chengqi Zhang. W ALL-e: World alignment by rule learning improves world model-based LLM agents,

  25. [33]

    Laird and Corey Clark

    Eli J. Laird and Corey Clark. On memory: A comparison of memory mechanisms in world models, 2025. URL https://arxiv.org/abs/2512.06983

    2025

  26. [34]

    MobileDreamer: Generative sketch world model for GUI agent, 2026

    Yilin Cao, Yufeng Zhong, Zhixiong Zeng, Liming Zheng, Jing Huang, Haibo Qiu, Peng Shi, Wenji Mao, and Wan Guanglu. MobileDreamer: Generative sketch world model for GUI agent, 2026. URL https://arxiv.org/abs/2601.04035

    2026

  27. [35]

    ToolVerse: Unlocking massive environments and long-horizon tasks for agentic reinforcement learning, 2026

    Shuaiyu Zhou, Fengpeng Yue, Zengjie Hu, Yuanzhe Shen, Chenyang Zhang, feng hong, Cao Liu, and Ke Zeng. ToolVerse: Unlocking massive environments and long-horizon tasks for agentic reinforcement learning, 2026. URL https://arxiv.org/abs/2607.15660

    2026 arXiv

  28. [36]

    AgentEvolver: Towards efficient self-evolving agent system, 2025

    Yunpeng Zhai, Shuchang Tao, Cheng Chen, Anni Zou, Ziqian Chen, Qingxu Fu, Shinji Mai, Li Yu, Jiaji Deng, Zouying Cao, Zhaoyang Liu, Bolin Ding, and Jingren Zhou. AgentEvolver: Towards efficient self-evolving agent system, 2025. URL https://arxiv.org/abs/2511.10395

    2025

  29. [37]

    Beyond experience re- trieval: Learning to generate utility-optimized structured experience for frozen LLMs

    Xuancheng Li, Haitao Li, Yujia Zhou, Yiqun Liu, and Qingyao Ai. Beyond experience re- trieval: Learning to generate utility-optimized structured experience for frozen LLMs. Open MIND, 2026. doi: 10.48550/arxiv.2602.02556. URL https://arxiv.org/abs/2602.02556

    2026 doi arXiv

  30. [38]

    Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang

    Nelson F. Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. Lost in the middle: How language models use long contexts. Transactions of the Association for Computational Linguistics , 12:157–173, 2024. doi: 10.1162/tacl\_a\ _0063...

    2024 arXiv

  31. [39]

    Cognitive workspace: Active memory management for LLMs – an empirical study of functional infinite context, 2025

    Tao An. Cognitive workspace: Active memory management for LLMs – an empirical study of functional infinite context, 2025. URL https://arxiv.org/abs/2508.13171

    2025 arXiv

  32. [40]

    Git context controller: Manage the context of LLM-based agents like git, 2025

    Junde Wu, Minhao Hu, Jiayuan Zhu, Jiazhen Pan, Yuyuan Liu, Min Xu, and Yueming Jin. Git context controller: Manage the context of LLM-based agents like git, 2025. URL https://arxiv.org/abs/2508.00031

    2025 arXiv

  33. [41]

    Scaling long-horizon LLM agent via context-folding, 2025

    Weiwei Sun, Miao Lu, Zhan Ling, Kang Liu, Xuesong Yao, Yiming Yang, and Jiecao Chen. Scaling long-horizon LLM agent via context-folding, 2025. URL https://arxiv.org/abs/2510 .11967

    2025

  34. [42]

    Inan, Lukas Wutschitz, Yanzhi Chen, Robert B

    Minki Kang, Weining Chen, Dongge Han, Huseyin A. Inan, Lukas Wutschitz, Yanzhi Chen, Robert B. Sim, and Saravan Rajmohan. ACON: Optimizing context compression for long- horizon LLM agents, 2025. URL https://arxiv.org/abs/2510.00615

    2025 arXiv

  35. [43]

    Diagnosing and mitigating context rot in long-horizon search, 2026

    Shijie Xia, Yikun Wang, Z Zhangqin Huang, and Pengfei Liu. Diagnosing and mitigating context rot in long-horizon search, 2026. URL https://arxiv.org/abs/2606.29718

    2026 arXiv

  36. [44]

    Plans don’t persist: Why context management is load bearing for LLM agents, 2026

    Aman Mehta and Anupam Datta. Plans don’t persist: Why context management is load bearing for LLM agents, 2026. URL https://arxiv.org/abs/2606.22953. 32

    2026 arXiv

  37. [45]

    Patil, Ion Stoica, and Joseph E

    Charles Packer, Sarah Wooders, Kevin Lin, Vivian Fang, Shishir G. Patil, Ion Stoica, and Joseph E. Gonzalez. MemGPT: Towards LLMs as operating systems, 2023. URL https: //arxiv.org/abs/2310.08560. arXiv preprint; not a peer-reviewed venue (despite being widely but incorrectly ...

    2023 arXiv

  38. [46]

    O’Brien, Carrie J

    Joon Sung Park, Joseph C. O’Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. Generative agents: Interactive simulacra of human behavior. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology (UIST), 2023. ...

    2023

  39. [47]

    A-MEM: Agentic memory for LLM agents, 2025

    Wujiang Xu, Zujie Liang, Kai Mei, Hang Gao, Juntao Tan, and Yongfeng Zhang. A-MEM: Agentic memory for LLM agents, 2025. URL https://arxiv.org/abs/2502.12110

    2025 arXiv

  40. [48]

    Graph-based agent memory: Taxonomy, techniques, and applications

    Chang Yang, Chuang Zhou, Yilin Xiao, Su Dong, Luyao Zhuang, Yujing Zhang, Zhu Wang, Zijin Hong, Zheng Yuan, Zhiyi Xiang, Shengyuan Chen, Huachi Zhou, Qinggang Zhang, Ninghao Liu, Jinsong Su, Bo An, Yi Chang, and Xiao Huang. Graph-based agent memory: Taxonomy, techniques, and a...

    2026 doi

  41. [49]

    WebCoach: Self-evolving web agents with cross-session memory guidance, 2025

    Genglin Liu, Shijie Geng, Sha Li, Hejie Cui, Sarah Zhang, Xin Liu, and Tianyi Liu. WebCoach: Self-evolving web agents with cross-session memory guidance, 2025. URL https://arxiv.org/ abs/2511.12997

    2025 arXiv

  42. [50]

    When stored evidence stops being usable: Scale-conditioned evaluation of agent memory, 2026

    Jiaqi Shao, Yiyi Lu, Yunzhen Zhang, and Bing Luo. When stored evidence stops being usable: Scale-conditioned evaluation of agent memory, 2026. URL https://arxiv.org/abs/2605.07313

    2026 arXiv

  43. [51]

    MINTEval: Evaluating memory under multi-target interference in long-horizon agent systems, 2026

    Hyunji Lee, Justin Chih-Yao Chen, Joykirat Singh, Zaid Khan, Elias Stengel-Eskin, and Mohit Bansal. MINTEval: Evaluating memory under multi-target interference in long-horizon agent systems, 2026. URL https://arxiv.org/abs/2605.18565

    2026 arXiv

  44. [52]

    FadeMem: Biologically-inspired forgetting for efficient agent memory

    Lei Wei, Xiao Peng, Xu Dong, Niantao Xie, and Bin Wang. FadeMem: Biologically-inspired forgetting for efficient agent memory. Open MIND , 2026. doi: 10.48550/arxiv.2601.18642. URL https://arxiv.org/abs/2601.18642

    2026 doi

  45. [53]

    MemPO: Self-memory policy optimization for long-horizon agents, 2026

    Ruoran Li, Xinghua Zhang, Haiyang Yu, Shitong Duan, Xiang Li, Wenxin Xiang, Chonghua Liao, Xudong Guo, Yongbin Li, and Jinli Suo. MemPO: Self-memory policy optimization for long-horizon agents, 2026. URL https://arxiv.org/abs/2603.00680

    2026 arXiv

  46. [54]

    Co-evolving LLM decision and skill bank agents for long-horizon tasks, 2026

    Xiyang Wu, Zongxia Li, Guangyao Shi, Alexander Duffy, Tyler Marques, Matthew Olson, Tianyi Zhou, and Dinesh Manocha. Co-evolving LLM decision and skill bank agents for long-horizon tasks, 2026. URL https://arxiv.org/abs/2604.20987

    2026 arXiv

  47. [55]

    Hijacking agent memory: Stealthy trojan attacks through conversational interaction, 2026

    Hongtao Wang, Se Yang, Yu Chen, and Puzhuo Liu. Hijacking agent memory: Stealthy trojan attacks through conversational interaction, 2026. URL https://arxiv.org/abs/2605.29960

    2026 arXiv

  48. [56]

    Forensic trajectory signatures for agent memory poisoning detection, 2026

    Jun Wen Leong. Forensic trajectory signatures for agent memory poisoning detection, 2026. URL https://arxiv.org/abs/2606.30566

    2026 arXiv

  49. [57]

    ReAct: Synergizing reasoning and acting in language models, 2022

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. ReAct: Synergizing reasoning and acting in language models, 2022. URL https: //arxiv.org/abs/2210.03629. ICLR 2023

    2022 arXiv

  50. [58]

    Del Rosario, Klaudia Krawiecka, and Christian Schroeder de Witt

    Ron F. Del Rosario, Klaudia Krawiecka, and Christian Schroeder de Witt. Architecting 33 resilient LLM agents: A guide to secure plan-then-execute implementations, 2025. URL https://arxiv.org/abs/2509.08646

    2025 arXiv

  51. [59]

    Toolformer: Language models can teach themselves to use tools

    Timo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu, Maria Lomeli, Eric Hambro, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. Toolformer: Language models can teach themselves to use tools. In Advances in Neural Information Processing Systems (NeurIPS), 2023....

    2023 arXiv

  52. [60]

    HuggingGPT: Solving AI tasks with ChatGPT and its friends in hugging face

    Yongliang Shen, Kaitao Song, Xu Tan, Dongsheng Li, Weiming Lu, and Yueting Zhuang. HuggingGPT: Solving AI tasks with ChatGPT and its friends in hugging face. In Advances in Neural Information Processing Systems (NeurIPS) , 2023. URL https://arxiv.org/abs/23 03.17580

    2023

  53. [61]

    iReDev: A knowledge-driven multi-agent framework for intelligent requirements development, 2025

    Dongming Jin, Weisong Sun, Jiangping Huang, Peng Liang, Jifeng Xuan, Yang Liu, and Zhi Jin. iReDev: A knowledge-driven multi-agent framework for intelligent requirements development, 2025. URL https://arxiv.org/abs/2507.13081

    2025 arXiv

  54. [62]

    SentiMM: A multimodal multi-agent framework for sentiment analysis in social media,

    Xilai Xu, Zilin Zhao, Chengye Song, Zining Wang, Jinhe Qiang, Jiongrui Yan, and Yuhuai Lin. SentiMM: A multimodal multi-agent framework for sentiment analysis in social media,

  55. [63]

    xoffense: An autonomous multi- agent framework for penetration testing with domain-adapted large language models, 2025

    Phung Duc Luong, Le Ngoc Bao, Nguyen Vu Khai Tam, Dong Huu Nguyen Khoa, Nguyen Huu Quyen, Van-Hau Pham, and Phan The Duy. xoffense: An autonomous multi- agent framework for penetration testing with domain-adapted large language models, 2025. URL https://arxiv.org/abs/2509.13021

    2025 arXiv

  56. [64]

    Lin, Eliot Krzysztof Jones, Donovan Julian Jasper, Ethan Ho, Anna H

    Justin W. Lin, Eliot Krzysztof Jones, Donovan Julian Jasper, Ethan Ho, Anna H. Wu, Arnold Tianyi Yang, Neil Perry, Andy Zou, Matt Fredrikson, J. Zico Kolter, Percy Liang, Dan Boneh, and Daniel E. Ho. Comparing AI agents to cybersecurity professionals in real-world penetration ...

    2025

  57. [65]

    Generative AI-driven hierarchical multi-agent framework for zero-touch optical networks, 2025

    Yao Zhang, Yuchen Song, Shengnan Li, Yan Shi, Shikui Shen, Xiongyan Tang, Min Zhang, and Danshi Wang. Generative AI-driven hierarchical multi-agent framework for zero-touch optical networks, 2025. URL https://arxiv.org/abs/2510.05625

    2025

  58. [66]

    Multi-agent LLM orchestration achieves deterministic, high-quality decision support for incident response, 2025

    Philip Drammeh. Multi-agent LLM orchestration achieves deterministic, high-quality decision support for incident response, 2025. URL https://arxiv.org/abs/2511.15755

    2025

  59. [67]

    Agent s2: A compositional generalist-specialist framework for computer use agents, 2025

    Saaket Agashe, Kyle Wong, Vincent Tu, Jiachen Yang, An-Ping Li, and Xin Eric Wang. Agent s2: A compositional generalist-specialist framework for computer use agents, 2025. URL https://arxiv.org/abs/2504.00906

    2025 arXiv

  60. [68]

    OmegaUse: Building a general-purpose GUI agent for autonomous task execution

    Le Zhang, Yixiong Xiao, Xinjiang Lu, Jingjia Cao, Yusai Zhao, Jingbo Zhou, Lang An, Zikan Feng, Wanxiang Sha, Yu Shi, Congxi Xiao, Jian Xiong, Yankai Zhang, Hua Wu, and Haifeng Wang. OmegaUse: Building a general-purpose GUI agent for autonomous task execution. Open MIND , 2026...

    2026 doi

  61. [69]

    Mohammed Mehedi Hasan, Hao Li, Emad Fallahzadeh, Gopi Krishnan Rajbahadur, Bram Adams, and Ahmed E. Hassan. An empirical study of testing practices in open source AI agent frameworks and agentic applications, 2025. URL https://arxiv.org/abs/2509.19185. 34

    2025 arXiv

  62. [70]

    A two-dimensional framework for AI agent design patterns: Cognitive function and execution topology, 2026

    Jia Huang and Joey Tianyi Zhou. A two-dimensional framework for AI agent design patterns: Cognitive function and execution topology, 2026. URL https://arxiv.org/abs/2605.13850

    2026 arXiv

  63. [71]

    Governing AI agents, 2025

    Noam Kolt. Governing AI agents, 2025. URL https://arxiv.org/abs/2501.07913

    2025 arXiv

  64. [72]

    Hunter, Carson Ezell, Emma Cabalé, Michael Gerovitch, Stewart Slocum, Kevin Wei, Nikola Jurkovic, Ariba Khan, Phillip J

    Stephen Casper, Luke Bailey, Robert C. Hunter, Carson Ezell, Emma Cabalé, Michael Gerovitch, Stewart Slocum, Kevin Wei, Nikola Jurkovic, Ariba Khan, Phillip J. K. Christof- fersen, A. Pinar Ozisik, Rakshit Trivedi, Dylan Hadfield-Menell, and Noam Kolt. The AI agent index, 2025...

    2025 arXiv

  65. [73]

    Reflexion: Language agents with verbal reinforcement learning

    Noah Shinn, Federico Cassano, Edward Berman, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. Reflexion: Language agents with verbal reinforcement learning. In Advances in Neural Information Processing Systems (NeurIPS) , 2023. URL https://arxiv.org/abs/2303.1 1366

    2023

  66. [74]

    Language models can perform single-utterance self-correction of perturbed reasoning, 2025

    Sam Silver, Jimin Sun, Ivan Zhang, Sara Hooker, and Eddie Kim. Language models can perform single-utterance self-correction of perturbed reasoning, 2025. URL https://arxiv.or g/abs/2506.15894

    2025 arXiv

  67. [75]

    Large language models can self-correct with key condition verification, 2024

    Zhenyu Wu, Qingkai Zeng, Zhihan Zhang, Zhaoxuan Tan, Chao Shen, and Meng Jiang. Large language models can self-correct with key condition verification, 2024. URL https: //arxiv.org/abs/2405.14092

    2024 arXiv

  68. [76]

    Decomposing LLM self-correction: The accuracy-correction paradox and error depth hypothesis, 2025

    Yin Li. Decomposing LLM self-correction: The accuracy-correction paradox and error depth hypothesis, 2025. URL https://arxiv.org/abs/2601.00828

    2025

  69. [77]

    CSC-SQL: Corrective self-consistency in text-to-SQL via reinforcement learning, 2025

    Lei Sheng and Shuaishuai Xu. CSC-SQL: Corrective self-consistency in text-to-SQL via reinforcement learning, 2025. URL https://arxiv.org/abs/2505.13271

    2025 arXiv

  70. [78]

    SHIELDA: Structured handling of exceptions in LLM-driven agentic workflows, 2025

    Jingwen Zhou, Jieshan Chen, Qinghua Lu, Dehai Zhao, and Liming Zhu. SHIELDA: Structured handling of exceptions in LLM-driven agentic workflows, 2025. URL https: //arxiv.org/abs/2508.07935

    2025 arXiv

  71. [79]

    Beyond entangled planning: Task-decoupled planning for long-horizon agents, 2026

    Yunfan Li, Bingbing Xu, Xueyun Tian, Xiucheng Xu, and Huawei Shen. Beyond entangled planning: Task-decoupled planning for long-horizon agents, 2026. URL https://arxiv.org/ab s/2601.07577

    2026

  72. [80]

    Voyager: An open-ended embodied agent with large language models, 2023

    Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. Voyager: An open-ended embodied agent with large language models, 2023. URL https://arxiv.org/abs/2305.16291. TMLR 2024

    2023 arXiv

  73. [81]

    ViReSkill: Vision-grounded replanning with skill memory for LLM-based planning in lifelong robot learning, 2025

    Tomoyuki Kagaya, Subramanian Lakshmi, Anbang Ye, Thong Jing Yuan, Jayashree Karlekar, Sugiri Pranata, Natsuki Murakami, Akira Kinose, and Yang You. ViReSkill: Vision-grounded replanning with skill memory for LLM-based planning in lifelong robot learning, 2025. URL https://arxi...

    2025

  74. [82]

    SEAgent: Self-evolving computer use agent with autonomous learning from experience, 2025

    Zeyi Sun, Ziyu Liu, Yuhang Zang, Yuhang Cao, Xiaoyi Dong, Tong Wu, Dahua Lin, and Jiaqi Wang. SEAgent: Self-evolving computer use agent with autonomous learning from experience, 2025. URL https://arxiv.org/abs/2508.04700

    2025 arXiv

  75. [83]

    Segment policy optimization: Effective segment-level credit assignment in RL for large language models, 2025

    Yiran Guo, Lijie Xu, Jie Liu, Dan Ye, and Shuang Qiu. Segment policy optimization: Effective segment-level credit assignment in RL for large language models, 2025. URL https://arxiv. org/abs/2505.23564. 35

    2025

  76. [84]

    Beyond trajectory-level attribution: Graph-based credit assignment for agentic reinforcement learn- ing, 2026

    Xin Cheng, Shuo He, Lang Feng, HaiYang Xu, Ming Yan, Lei Feng, and Bo An. Beyond trajectory-level attribution: Graph-based credit assignment for agentic reinforcement learn- ing, 2026. URL https://arxiv.org/abs/2605.26684

    2026 arXiv

  77. [85]

    Beyond uniform credit assignment: Selective eligibility traces for RL VR, 2026

    Chaoli Mou, Zhan Zhuang, Xinning Chen, and Yu Zhang. Beyond uniform credit assignment: Selective eligibility traces for RL VR, 2026. URL https://arxiv.org/abs/2605.05965

    2026 arXiv

  78. [86]

    MICA: Multi-granularity intertemporal credit assignment for long-horizon emotional support dialogue

    Naifan Zhang, Ruihan Sun, Jinwei Su, Hengjie Yang, Zhengyuan Pan, Zhaohan Chen, and Xiaofan Zhang. MICA: Multi-granularity intertemporal credit assignment for long-horizon emotional support dialogue. Open MIND , 2026. doi: 10.48550/arxiv.2603.06194. URL https://arxiv.org/abs/2...

    2026 doi arXiv

  79. [87]

    Memory-r2: Fair credit assignment for long-horizon memory- augmented LLM agents, 2026

    Sikuan Yan, Ahmed Bahloul, Ercong Nie, Susanna Schwarzmann, Riccardo Trivisonno, Volker Tresp, and Yunpu Ma. Memory-r2: Fair credit assignment for long-horizon memory- augmented LLM agents, 2026. URL https://arxiv.org/abs/2605.21768

    2026 arXiv

  80. [88]

    Demystifying reinforcement learning for long-horizon tool-using agents: A comprehensive recipe, 2026

    Xixi Wu, Qianguo Sun, Ruiyang Zhang, Chao Song, Junlong Wu, Yiyan Qi, and Hong Cheng. Demystifying reinforcement learning for long-horizon tool-using agents: A comprehensive recipe, 2026. URL https://arxiv.org/abs/2603.21972

    2026

  81. [89]

    Let’s verify step by step, 2023

    Hunter Lightman, Vineet Kosaraju, Yura Burda, Harri Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. Let’s verify step by step, 2023. URL https://arxiv.org/abs/2305.20050. ICLR 2024

    2023 arXiv

  82. [90]

    Entropy-regularized process reward model, 2024

    Hanning Zhang, Pengcheng Wang, Shizhe Diao, Yong Lin, Rui Pan, Hanze Dong, Dylan Zhang, Pavlo Molchanov, and Tong Zhang. Entropy-regularized process reward model, 2024. URL https://arxiv.org/abs/2412.11006

    2024

  83. [91]

    GroundedPRM: Tree-guided and fidelity-aware process reward mod- eling for step-level reasoning, 2025

    Yao Zhang, Yu Wu, Haowei Zhang, Weiguo Li, Haokun Chen, Jingpei Wu, Guohao Li, Zhen Han, and Volker Tresp. GroundedPRM: Tree-guided and fidelity-aware process reward mod- eling for step-level reasoning, 2025. URL https://arxiv.org/abs/2510.14942

    2025

  84. [92]

    GRPO is secretly a process reward model, 2025

    Michael Sullivan and Alexander Koller. GRPO is secretly a process reward model, 2025. URL https://arxiv.org/abs/2509.21154

    2025 arXiv

  85. [93]

    Think-RM: Enabling long-horizon reasoning in generative reward models, 2025

    Ilgee Hong, Changlong Yu, Liang Qiu, Weixiang Yan, Zhenghao Xu, Haoming Jiang, Qingru Zhang, Qin Lu, Xin Liu, Chao Zhang, and Tuo Zhao. Think-RM: Enabling long-horizon reasoning in generative reward models, 2025. URL https://arxiv.org/abs/2505.16265

    2025 arXiv

  86. [94]

    AgentPRM: Process reward models for LLM agents via step-wise promise and progress, 2025

    Zhiheng Xi, Chenyang Liao, Guanyu Li, Yajie Yang, Wenxiang Chen, Zhihao Zhang, Binghai Wang, Senjie Jin, Yuhao Zhou, Jian Guan, Wei Wu, Tao Ji, Tao Gui, Qi Zhang, and Xuan- jing Huang. AgentPRM: Process reward models for LLM agents via step-wise promise and progress, 2025. URL...

    2025

  87. [95]

    SWE-TRACE: Optimizing long-horizon SWE agents through rubric process reward models and heuristic test-time scaling, 2026

    Hao Han, Jin Xie, Xuehao Ma, Weiquan Zhu, Ziyao Zhang, Zhiliang Long, Hongkai Chen, and Qingwen Ye. SWE-TRACE: Optimizing long-horizon SWE agents through rubric process reward models and heuristic test-time scaling, 2026. URL https://arxiv.org/abs/2604.14820

    2026 arXiv

  88. [96]

    Agentic rein- forcement learning for search misaligns instruction-tuning, 2025

    Yushi Yang, Shreyansh Padarha, Sarah Ball, Andrew Lee, and Adam Mahdi. Agentic rein- forcement learning for search misaligns instruction-tuning, 2025. URL https://arxiv.org/ab s/2510.17431

    2025

  89. [97]

    Self-evolving LLM agents with in-distribution optimization, 2026

    Yudi Zhang, Meng Fang, Zhenfang Chen, and Mykola Pechenizkiy. Self-evolving LLM agents with in-distribution optimization, 2026. URL https://arxiv.org/abs/2606.07367. 36

    2026 arXiv

  90. [98]

    Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan

    Carlos E. Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan. SWE-bench: Can language models resolve real-world GitHub issues?,

  91. [99]

    Qi, Hao Yu, Lei Yu, Dezhi Ran, Muhan Zeng, Bo Shen, Pan Bian, Guangtai Liang, Bei Guan, Pengjie Huang, Tao Xie, Yongji Wang, and Qianxiang Wang

    Daoguang Zan, Zhirong Huang, Ailun Yu, Shaoxin Lin, Yifan Shi, Wei Liu, Dong Chen, Z. Qi, Hao Yu, Lei Yu, Dezhi Ran, Muhan Zeng, Bo Shen, Pan Bian, Guangtai Liang, Bei Guan, Pengjie Huang, Tao Xie, Yongji Wang, and Qianxiang Wang. SWE-bench-java: A GitHub issue resolving bench...

    2024 arXiv

  92. [100]

    Multi-SWE-bench: A multilingual benchmark for issue resolving, 2025

    Daoguang Zan, Zhirong Huang, Wei Liu, Hanwu Chen, Linhao Zhang, Shulin Xin, Chen Lü, Qi Liu, Xiaojian Zhong, Aoyan Li, Siyao Liu, Yong-Sheng Xiao, Liangqiang Chen, Yuyu Zhang, Jing Su, Tianyu Liu, Rui Long, Kai Shen, and Xiang Liang. Multi-SWE-bench: A multilingual benchmark f...

    2025 arXiv

  93. [101]

    Lieret, Joyce Yang, Xindi Wu, Ori Press, Niklas Muennighoff, Gabriel Synnaeve, Karthik Narasimhan, Diyi Yang, Sida I

    John Yang, Carlos Jimenez-Gomez, Alex Zhang, K. Lieret, Joyce Yang, Xindi Wu, Ori Press, Niklas Muennighoff, Gabriel Synnaeve, Karthik Narasimhan, Diyi Yang, Sida I. Wang, and Ofir Press. SWE-bench multimodal: Do AI systems generalize to visual software domains?,

  94. [102]

    Garg, Niklas Lauffer, Andrew Park, Nitin Pasari, Chetan Rane, K Sampath, Maya Krishnan, Srivatsa Kundurthy, Sean M

    Xiang Deng, Jeff Da, Edwin Pan, Yun He, Charles Ide, K. Garg, Niklas Lauffer, Andrew Park, Nitin Pasari, Chetan Rane, K Sampath, Maya Krishnan, Srivatsa Kundurthy, Sean M. Hendryx, Zifan Wang, Vijay Bharadwaj, Jeff Holm, Raja Aluri, Chen Bo Calvin Zhang, Noah Jacobson, Bing Li...

    2025 arXiv

  95. [103]

    SWE-bench+: Enhanced coding benchmark for LLMs, 2024

    Reem Aleithan, Huayuan Xue, Mohammad Mahdi Mohajer, Elijah Nnorom, Gias Uddin, and Song Wang. SWE-bench+: Enhanced coding benchmark for LLMs, 2024. URL https: //arxiv.org/abs/2410.06992

    2024 arXiv

  96. [104]

    Zadorozhny, Ivan Lopatin, Dmitry Babayev, Alena Fenogenova, and Valentin Malykh

    Pavel Adamenko, Mikhail Ivanov, Aidar Valeev, Rodion Levichev, P. Zadorozhny, Ivan Lopatin, Dmitry Babayev, Alena Fenogenova, and Valentin Malykh. SWE-MERA: A dy- namic benchmark for agenticly evaluating large language models on software engineering tasks, 2025. URL https://ar...

    2025 arXiv

  97. [105]

    URL https://arxiv.org/abs/2410.03859

    arXiv

  98. [106]

    Dissecting the SWE-bench leaderboards: Profiling submitters and architectures of LLM- and agent-based repair systems, 2025

    Matías Martínez and Xavier Franch. Dissecting the SWE-bench leaderboards: Profiling submitters and architectures of LLM- and agent-based repair systems, 2025. URL https: //arxiv.org/abs/2506.17208

    2025

  99. [107]

    What’s in a benchmark? the case of SWE-bench in automated program repair

    Matías Martínez and Xavier Franch. What’s in a benchmark? the case of SWE-bench in automated program repair. Open MIND , 2026. doi: 10.48550/arxiv.2602.04449. URL https://arxiv.org/abs/2602.04449

    2026 doi

  100. [108]

    Coherence collapse: Diagnosing why code agents fail after reaching the right code, 2026

    Myeongsoo Kim, Dingmin Wang, Siwei Cui, Farima Farmahinifarahani, Terry Yue Zhuo, Shweta Garg, Baishakhi Ray, Rajdeep Mukherjee, and Varun Kumar. Coherence collapse: Diagnosing why code agents fail after reaching the right code, 2026. URL https://arxiv.org/ abs/2603.24631

    2026 arXiv

  101. [109]

    Does SWE-bench- verified test agent ability or model memory?, 2025

    Thanosan Prathifkumar, Noble Saji Mathews, and Meiyappan Nagappan. Does SWE-bench- verified test agent ability or model memory?, 2025. URL https://arxiv.org/abs/2512.10218

    2025

  102. [110]

    Xu, Hao Zhu, Xuhui Zhou, Robert Lo, Abishek Sridhar, Xianyi Cheng, Tianyue Ou, Yonatan Bisk, Daniel Fried, Uri Alon, and Graham Neubig

    Shuyan Zhou, Frank F. Xu, Hao Zhu, Xuhui Zhou, Robert Lo, Abishek Sridhar, Xianyi Cheng, Tianyue Ou, Yonatan Bisk, Daniel Fried, Uri Alon, and Graham Neubig. WebArena: A realistic web environment for building autonomous agents, 2023. URL https://arxiv.org/ abs/2307.13854. ICLR 2024

    2023 arXiv

  103. [111]

    OSWorld: Benchmarking multimodal agents for open-ended tasks in real computer environments

    Tianbao Xie, Danyang Zhang, Jixuan Chen, Xiaochuan Li, Siheng Zhao, Ruisheng Cao, Toh Jing Hua, Zhoujun Cheng, Dongchan Shin, Fangyu Lei, Yitao Liu, Yiheng Xu, Shuyan Zhou, Silvio Savarese, Caiming Xiong, Victor Zhong, and Tao Yu. OSWorld: Benchmarking multimodal agents for op...

    2024 arXiv

  104. [112]

    AgentBench: Evaluating LLMs as agents, 2023

    Xiao Liu, Hao Yu, Hanchen Zhang, Yifan Xu, Xuanyu Lei, Hanyu Lai, Yu Gu, Hangliang Ding, Kaiwen Men, Kejuan Yang, Shudan Zhang, Xiang Deng, Aohan Zeng, Zhengxiao Du, Chenhui Zhang, Sheng Shen, Tianjun Zhang, Yu Su, Huan Sun, Minlie Huang, Yuxiao Dong, and Jie Tang. AgentBench:...

    2023 arXiv

  105. [113]

    𝜏 -bench: A benchmark for tool-agent-user interaction in real-world domains, 2024

    Shunyu Yao, Noah Shinn, Pedram Razavi, and Karthik Narasimhan. 𝜏 -bench: A benchmark for tool-agent-user interaction in real-world domains, 2024. URL https://arxiv.org/abs/2406 .12045. ICLR 2025. 37

    2024

  106. [114]

    Zico Kolter, Nicolas Flammar- ion, and Maksym Andriushchenko

    Thomas Kuntz, Agatha Duzan, Hao Zhao, Francesco Croce, J. Zico Kolter, Nicolas Flammar- ion, and Maksym Andriushchenko. OS-harm: A benchmark for measuring safety of computer use agents, 2025. URL https://arxiv.org/abs/2506.14866

    2025

  107. [115]

    AndroidControl-curated: Revealing the true potential of GUI agents through benchmark purification, 2025

    Ho Fai Leung, Xiaoyan Xi, and Fei Zuo. AndroidControl-curated: Revealing the true potential of GUI agents through benchmark purification, 2025. URL https://arxiv.org/abs/2510.18488

    2025

  108. [116]

    Li, and Xin Eric Wang

    Gonzalo Gonzalez-Pumariega, Saaket Agashe, Jiachen Yang, A. Li, and Xin Eric Wang. On the reliability of computer use agents, 2026. URL https://arxiv.org/abs/2604.17849

    2026 arXiv

  109. [117]

    GAIA: a benchmark for general AI assistants, 2023

    Grégoire Mialon, Clémentine Fourrier, Craig Swift, Thomas Wolf, Yann LeCun, and Thomas Scialom. GAIA: a benchmark for general AI assistants, 2023. URL https://arxiv.org/abs/23 11.12983. ICLR 2024

    2023

  110. [118]

    WindowsWorld: A process-centric benchmark of autonomous GUI agents in professional cross- application environments, 2026

    Jinchao Li, Yunxin Li, Chenrui Zhao, Zhenran Xu, Baotian Hu, and Min Zhang (111999). WindowsWorld: A process-centric benchmark of autonomous GUI agents in professional cross- application environments, 2026. URL https://arxiv.org/abs/2604.27776

    2026 arXiv

  111. [119]

    SWE-bench-CL: Continual learning for coding agents, 2025

    Thomas Joshi, Shayan Chowdhury, and Fatih Uysal. SWE-bench-CL: Continual learning for coding agents, 2025. URL https://arxiv.org/abs/2507.00014

    2025 arXiv

  112. [120]

    Building self-evolving agents via experience-driven lifelong learning: A framework and benchmark, 2025

    Yuxuan Cai, Yipeng Hao, Jie Zhou, Hang Yan, Zhikai Lei, Rui Zhen, Zhenhua Han, Yutao Yang, Junsong Li, Qianjun Pan, Tianyu Huai, Qin Chen, Xin Li, Kai Chen, Bo Zhang, Xipeng Qiu, and Liang He. Building self-evolving agents via experience-driven lifelong learning: A framework a...

    2025

  113. [121]

    macOSWorld: A multilingual interactive benchmark for GUI agents, 2025

    Pei Yang, Hai Ci, and Mike Zheng Shou. macOSWorld: A multilingual interactive benchmark for GUI agents, 2025. URL https://arxiv.org/abs/2506.04135

    2025

  114. [122]

    Strained coherence: A pre-failure signal in coding agent execution trajectories, 2026

    Marut Pandya, Kasey Zhang, and Baiqing Lyu. Strained coherence: A pre-failure signal in coding agent execution trajectories, 2026. URL https://arxiv.org/abs/2606.07889. 38

    2026 arXiv

  115. [123]

    Vending-bench: A benchmark for long-term coherence of autonomous agents, 2025

    Axel Backlund and Lukas Petersson. Vending-bench: A benchmark for long-term coherence of autonomous agents, 2025. URL https://arxiv.org/abs/2502.15840

    2025 arXiv

  116. [124]

    Technical report: Evaluating goal drift in language model agents, 2025

    Rauno Arike, Elizabeth Donoway, Henning Bartsch, and Marius Hobbhahn. Technical report: Evaluating goal drift in language model agents, 2025. URL https://arxiv.org/abs/2505.02709

    2025 arXiv

  117. [125]

    EvoAgentBench: Benchmarking agent self-evolution via ability transfer, 2026

    Xingze Gao, Chuanrui Hu, Hongda Chen, Pengfei Yao, Zhao Wang, Yi Bai, Z Z Wu, Yunyun Han, Xiaofeng Cong, Jie Gui, Yafeng Deng, and Teng Li. EvoAgentBench: Benchmarking agent self-evolution via ability transfer, 2026. URL https://arxiv.org/abs/2607.05202

    2026 arXiv

  118. [126]

    The coming crisis of multi-agent misalignment: AI alignment must be a dynamic and social process, 2025

    Florian Carichon, Aditi Khandelwal, Marylou Fauchard, and Golnoosh Farnadi. The coming crisis of multi-agent misalignment: AI alignment must be a dynamic and social process, 2025. URL https://arxiv.org/abs/2506.01080

    2025 arXiv

  119. [127]

    A sober look at agentic misalignment in automated workflows, 2026

    Wenqian Ye, Bo Yuan, Zhichao Xu, Ye Tian, Yawei Wang, Henry Kautz, and Aidong Zhang. A sober look at agentic misalignment in automated workflows, 2026. URL https://arxiv.org/ abs/2605.24197

    2026 arXiv

  120. [128]

    Cheap talk, empty promise: Frontier LLMs easily break public promises for self-interest, 2026

    Jerick Shi, Terry Jingcheng Zhang, Zhijing Jin, and Vincent Conitzer. Cheap talk, empty promise: Frontier LLMs easily break public promises for self-interest, 2026. URL https: //arxiv.org/abs/2604.04782

    2026 arXiv

  121. [129]

    Governance decay: How context compaction silently erases safety constraints in long-horizon LLM agents, 2026

    Shiyang Chen. Governance decay: How context compaction silently erases safety constraints in long-horizon LLM agents, 2026. URL https://arxiv.org/abs/2606.22528

    2026 arXiv

  122. [130]

    Tiered agentic oversight: A hierarchical multi-agent system for healthcare safety, 2025

    Yubin Kim, Hyewon Jeong, Chanwoo Park, Eugene Park, Haipeng Zhang, Xin Liu, Hyeon- hoon Lee, Daniel McDuff, Marzyeh Ghassemi, Cynthia Breazeal, Samir Tulebaev, and Hae Won Park. Tiered agentic oversight: A hierarchical multi-agent system for healthcare safety, 2025. URL https:...

    2025

  123. [131]

    Ensemble monitoring for AI control: Diverse signals outweigh more compute, 2026

    Eugene Koran, Yejun Yun, Samantha Tetef, Benjamin Arnav, and Pablo Bernabeu-Pérez. Ensemble monitoring for AI control: Diverse signals outweigh more compute, 2026. URL https://arxiv.org/abs/2605.15377

    2026 arXiv

  124. [132]

    Managed autonomy at runtime: Gear-based safety and governance for single- and multi-agent cyber-physical systems, 2026

    Srini Ramaswamy and Wang Miaosheng. Managed autonomy at runtime: Gear-based safety and governance for single- and multi-agent cyber-physical systems, 2026. URL https://arxi v.org/abs/2607.00334

    2026 arXiv

  125. [133]

    Kim, Vasu Sharma, Sean O’Brien, and Kevin Zhu

    Mrinal Agarwal, Saad Rana, Theo Sundoro, Hermela Berhe, S. Kim, Vasu Sharma, Sean O’Brien, and Kevin Zhu. WOLF: Werewolf-based observations for LLM deception and false- hoods, 2025. URL https://arxiv.org/abs/2512.09187

    2025

  126. [134]

    SALT: Step-level advantage assignment for long-horizon agents via trajectory graph, 2025

    Jiazheng Li, Yawei Wang, David Yan, Yijun Tian, Zhichao Xu, Huan Song, Panpan Xu, and Lin Lee Cheong. SALT: Step-level advantage assignment for long-horizon agents via trajectory graph, 2025. URL https://arxiv.org/abs/2510.20022

    2025

  127. [135]

    MemWeaver: Weaving hybrid memories for traceable long-horizon agentic reasoning,

    Juexiang Ye, Xue Li, Xinyu Yang, Chengkai Huang, Lanshun Nie, Lina Yao, and Dechen Zhan. MemWeaver: Weaving hybrid memories for traceable long-horizon agentic reasoning,

  128. [136]

    Slipstream: Trajectory-grounded compaction validation for long-horizon agents, 2026

    Zhuofu Chen, Rui Pan, Yinwei Dai, and Ravi Netravali. Slipstream: Trajectory-grounded compaction validation for long-horizon agents, 2026. URL https://arxiv.org/abs/2605.08580. 39

    2026 arXiv

  129. [137]

    Filtering harmful actions isn’t enough: Phantom transfer in agentic SDF,

    Chinmayi Dixit. Filtering harmful actions isn’t enough: Phantom transfer in agentic SDF,

  130. [141]

    URL https://arxiv.org/abs/2601.18204

  131. [2023]

    ICLR 2024

    URL https://arxiv.org/abs/2310.06770. ICLR 2024

    2024 arXiv

  132. [2024]

    URL https://arxiv.org/abs/2410.07484

    arXiv

  133. [2025]

    URL https://arxiv.org/abs/2508.18108

    arXiv

  134. [2026]

    URL https://arxiv.org/abs/2607.10750

    arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.