ZOTPAPER

Research agents can reuse experience to avoid unnecessary experiments

Wei and colleagues introduce Experimental Experience Modeling, a memory framework that helps autonomous research agents decide whether a candidate direction merits costly full-scale evaluation. Instead of treating every experiment independently, the framework distills earlier outcomes into reusable records, retrieves relevant evidence, and commissions a low-cost pilot when the available evidence is insufficient.

Linked story artifacts help authors supervise narrative-generating agents

Wang et al. built an authoring environment that represents an interactive narrative as connected outlines, worldbuilding materials, narrative graphs and executable rules rather than scattered agent-generated files. In a 12-participant within-subject study, authors found it easier to formulate revisions, inspect changes and understand story structure with NarrativeSteward than with general-purpose agents.

Recursive testing loops improve generated games across two engines

Wu and colleagues present RSIGame, an autonomous development framework designed to improve generated games without repeatedly optimizing against the same narrow tests. Across 140 GameCraft-Bench tasks, five generators and the Godot and Phaser engines, the authors report consistent quality gains under matched development budgets.

Tailoring agent harnesses per task beats one-size-fits-all optimization

Zhang et al. introduce a framework that converts artifacts from a completed harness search into a structured playbook, then trains an editor to patch the resulting global harness for individual instances. Across seven benchmarks covering interactive agents, software engineering and long-horizon terminal work, the authors report that these tailored harnesses consistently outperform global-harness baselines.

Dataset turns failed agent traces into diagnosis and repair supervision

Zhu et al. present the Agent Error Dataset, a collection that preserves failed agent trajectories, identifies likely decision errors, and proposes evidence-checked corrections. Matched replay experiments suggest these corrections have practical value, while initial fine-tuning results show that the data can improve both failure diagnosis and action recovery.

Self-programmed memory adapts retrieval depth to each agent query

Wang and Liu present MemCodex, an agent memory architecture that can rewrite how different forms of experience are constructed, indexed, retrieved and routed. Across the reported evaluation, it raised average task success by 10.1% relative to the strongest adaptive-memory baseline while using fewer context tokens and reducing inference time.

Research-guided software evolution improves fixed-model agents on two benchmarks

Yang et al. introduce ScholarEvolve, a framework for improving the software harness that controls an agent’s tools, context, memory, skills, and execution workflow. On AppWorld and Tau2-Bench, research-guided harness changes substantially improved task completion while the underlying language models remained fixed.

Persistent context graphs cut agent memory costs without extra model calls

Yang et al. introduce ReCAP, a memory system for long-running LLM agents that records which earlier messages mattered and how they depended on one another. Across coding-agent evaluations, it substantially reduced context and estimated restoration latency while maintaining comparable task quality, and sometimes outperformed supplying the complete history.

Multi-agent orchestration produced proofs for five open research problems

Cai et al. present Cogentic, a harness designed to sustain mathematical research across many model calls rather than relying on a single response. Starting from problem statements without expert hints, the system generated claimed novel results for five open problems in online learning, auction theory, and mechanism design, which the authors report were independently checked by domain experts.

Contrasting multiple trajectories sharpens skill learning for LLMs

Li et al. propose SkillCome, a method that improves large language models by iteratively refining textual skills. For each question, the system generates multiple trajectories, contrasts successful and failed ones to pinpoint effective actions, and accumulates evidence across iterations in a dual memory system. On six benchmarks spanning reasoning and agentic tasks, SkillCome consistently outperformed prior skill evolution baselines, with gains up to +5.69 points.

More Harness Programs Do Not Necessarily Produce Useful Specialization

Xu and colleagues test whether generated LLM harnesses, programs that structure model inference, offer genuine task specialization rather than extra chances to obtain a correct answer. On MATH-500, the generated programs produced few stable advantages over a baseline, enabled no gain from advance selection, and tied identical baseline copies under a matched execution budget.

Refining rejected candidates makes AI system optimization more efficient

Lyu and colleagues test an alternative to optimization loops that simply discard candidates failing an acceptance threshold. Across three benchmarks, their method produced stronger task results with fewer evaluations by treating rejected configurations as intermediate steps rather than dead ends.

StepLearn lets agents use new knowledge immediately while requiring validation before persistent reuse.

The authors introduce StepLearn, a nonparametric framework for test-time learning in LLM agents. Instead of learning only from completed episodes, StepLearn distills individual transitions into hypotheses that can guide the next decision, while deferring persistent reuse until those hypotheses are prospectively validated in later episodes. Across five rounds on WebArena-Lite and ALFWorld, StepLearn outperforms the strongest baseline EvoTest by 2.2 to 12.7 percentage points.

Typed action encoding helps language models learn adaptable agent policies

Zhan et al. augment pretrained language models with an action encoder that directly scores the structured actions available in an environment, rather than generating each action token by token. The authors report that this architecture improves interactive task performance through reinforcement learning, including when the underlying language model remains frozen.

Statistical gates help self-editing agents avoid lasting performance regressions

Wang et al. examine how agents decide whether to retain proposed edits to persistent instructions that encode workflows, tool use and decision rules. Their SAGE gate compares old and edited skills on the same validation items, penalizes regressions, and abstains when evidence of improvement is statistically weak. Across five benchmarks and four language models, it reduced regressions in 19 of 20 settings while producing the highest final score in every setting.

Targeted feedback improves memory retrieval for language model agents

Liang and colleagues introduce UpliftMem, a retrieval method that scores an entire set of memories by its execution uplift relative to running the same frozen agent without memory. Across three agent benchmarks, the method produced the highest success rates among the evaluated baselines while requiring no additional probing at test time.

Whole-system training improves general-purpose agents’ use of external tools

Mao and colleagues present WEFT, a framework for scaling tool-use post-training across the entire agent interaction system rather than generating more executable environments alone. Across several models and benchmarks, WEFT outperformed matched-size environment-scaling baselines, with the largest reported gain reaching 12.27 percentage points.

Entity maps help search agents connect evidence across documents

Jeong et al. introduce CorpusMap, an offline-built navigation layer for agents searching large, heterogeneous document collections. Across three benchmarks and seven language models, agents using the map found distributed evidence and answered questions more effectively than agents searching a flat corpus, while consuming fewer tokens on average.

Adaptive mechanism selection improves LLM optimization across changing search states

Wei and colleagues present OptiCom, a framework for dynamically composing the mechanisms used by LLM-driven optimizers rather than relying on one fixed search configuration. Across 32 benchmark groups, OptiCom ranked first by maximum score in 23 groups and achieved an average rank of 1.72 among 14 evaluated configurations.

Search agents improve by deciding when to compress and reconsider

Ma et al. present a web-search harness that lets an agent define success criteria, seek an answer, verify it and decide whether to stop or redirect its search. Their 35-billion-parameter model scored 72.83 on BrowseComp and improved over its base model across several multilingual, investigative and product-search evaluations. The authors attribute part of these gains to learned, autonomous context compression.

Incorrect peer messages can override an agent’s evidence-backed answer

Gong and colleagues isolate the effect of message passing in multi-agent LLM systems by holding a downstream agent’s task and evidence fixed while varying the upstream message. They find that communication can correct an initially wrong receiver, but an incorrect message can overturn an initially correct answer in up to 32% of cases, usually by inducing the message’s specific error.

Validation-gated co-evolution improves agent models and execution harnesses

Qi and colleagues introduce VACE, a training loop that alternates updates to a language model’s weights with revisions to its harness, the instructions, skills and execution rules surrounding the model. Using Qwen3.5-9B, they report that accepting harness changes only after validation produced better results than weight-only reinforcement learning or alternation without a validation gate.

Execution graph helps off-the-shelf teachers supervise student agents better

The authors propose GC-OPD, a distillation method that gives an off-the-shelf teacher access to a graph of its own past executions when scoring student trajectories. Across ScienceWorld, ALFWorld Unseen and WebShop, it improved mean success over vanilla on-policy distillation, and on ScienceWorld and ALFWorld it also surpassed OPD baselines using GRPO-trained teachers, without retraining the teacher.

Specialized executors follow declared agent plans more faithfully

Oota et al. separate agent planning into two problems: choosing an appropriate planning mode and executing that mode faithfully. Across four benchmarks and three language models, generic Plan+ReAct agents frequently abandoned their declared structure, while pattern-specific executors largely closed this execution gap. The remaining weakness was selection, since models did not reliably choose the best mode for each task.

Counterfactual credit teaches language models which memories stay useful

Tang, Liu and Sarabi address a central problem in persistent language-model memory: a rewrite may help only much later, and its apparent value may actually come from information already stored. Their Memory Gain Policy Optimization method compares updated and prior memory states to isolate the value created by each rewrite, improving document-level information extraction while cutting average memory length by nearly 80% from the initial policy.

FOCUS Compresses Agent Histories While Preserving Future Decisions

Dixit et al. introduce FOCUS, a training-free layer that removes parts of an agent’s interaction history according to their estimated influence on future decisions. Across tool use, question answering, web navigation and multi-turn dialogue benchmarks, the authors report lower context requirements and, in some settings, better task success than retaining the full history.

Delayed recommendation feedback can guide useful evolution of agent memory

Mao et al. present TIDE, a framework that evolves an agent’s external memory using delayed signals such as clicks, conversions, and negative feedback while leaving the underlying generative model fixed. In e-commerce content generation, the approach improved performance in temporal replay and an online A/B test, while also outperforming comparison methods on a delayed-label benchmark.

Retrieving existing agent skills improves optimization across different execution harnesses

Chu et al. introduce Retrieval-Augmented Skill Optimization, or RASO, for improving the natural-language instructions that guide agents inside particular tool and execution environments. Across four benchmarks and two language models, the authors report that retrieving and adapting existing skills consistently outperformed optimization baselines that lacked retrieval during initialization and updating.

Agents Learn Reusable Control Code Instead of Rebuilding Context

Li et al. train an agent harness from a controller-free scaffold using execution failures, localized code repairs and held-out validation. Across two web-agent benchmarks and models ranging from 4B to 120B parameters, the resulting harness matched or exceeded competing methods in most settings while cutting model calls by up to 91.8%.

Self-distillation trains GUI agents for longer, memory-dependent tasks

Zhang et al. extend on-policy self-distillation from selecting interface elements to completing multi-turn GUI tasks. Their two-stage method first teaches the model to use corrective guidance, then distills step-specific reasoning and task-relevant memories into the agent, improving success rates on AndroidWorld and MobileWorld.

Workplace AI teammates disrupt tacit norms, trust, and human agency

Qadri and colleagues studied an autonomous agent embedded across several teams at a large technology company, using semi-structured interviews to examine its role in everyday collaborative work. They report that adoption depended not only on technical capability, but also on whether the agent respected unwritten communication norms and allowed people to retain meaningful control.

Developers write detailed, evolving instructions for agentic workflows but often skip safety measures

The authors analyzed 1,248 Markdown-based GitHub Agentic Workflow files across 276 repositories. They found that workflows contain a median of 556.5 words and are maintained for months, yet only 9.4% address prompt-injection attacks. The study also created a taxonomy of instruction types and evaluated LLM-based labeling.

Compiling agent skills into state machines improves task success

Li presents an incremental compiler that converts written agent skills into extended finite state machines, which track data and execution progress while enforcing transitions between operations. Across four benchmarks and four executors, HEXIS improved success over a Skill + ReAct baseline by an average of 16.2 percentage points.

Domain constraints can guide the design of context for AI

Xu et al. propose Constraint-Driven Context Engineering, a method for turning domain constraints into structured interfaces between AI systems and their operating environments. Applied to educational assessment, healthcare decision support and financial-distress prediction, the method distinguishes constraints that guide model behaviour, enforce boundaries or help verify outputs.

Structuring noisy file environments improves agents’ performance on complex tasks

Wu et al. developed Env-Rethink, a framework built around a post-trained 27-billion-parameter model that prepares persistent file environments for LLM agents. Across 30 tasks and nine models, the authors report a 15.1 percentage-point increase in mean rubric pass rate after environment preparation.

Entity-structured indexes help agents retrieve precise long-term memories

Meng et al. built a long-term agent memory architecture that partitions interactions into thematic episodes and indexes facts by entity, type, property, value, time, and provenance. On the LoCoMo and LongMemEval benchmarks, the authors report high answer accuracy with smaller, evidence-focused contexts rather than full histories or lossy summaries.

Verified memories help chip-design agents transfer lessons across tasks

AlRabah et al. developed a memory layer that lets EDA agents reuse procedures only after the resulting chip design artifacts pass synthesis, simulation, or formal checks. Under matched model and tool settings, the system improved equivalence-passing RTL rewrites and transferred a frozen skill library to unseen verification tasks.

Adaptively choosing how to revise LLM skills outperforms fixed revision strategies

The authors propose StraTune, a method that adaptively selects which revision operator to apply when evolving textual skills for LLMs. By maintaining an optimization state of past operator outcomes and letting a frozen optimizer LLM choose the operator each round, StraTune outperforms five baselines on four benchmarks. Gains are attributed to the adaptive selection, as fixed, random, scheduled, and bandit strategies all score lower.

Reusable plugins let agent harness improvements accumulate across tasks

Shi et al. replace whole-program harness rewriting with an optimization loop built around reusable plugins, which encapsulate mechanisms such as roles, skills, memory and tools. The authors report better results than existing harness optimization methods across software engineering, terminal interaction and question answering, with gains that transfer to other solver models.

Fine-Grained Training Data Improves Agents on Interactive CAPTCHA Tasks

Zhang et al. built CaptchaArena, a dataset of interactive CAPTCHA tasks whose reference solutions were verified by execution in a browser. Their CaptchaAgent reached 70.5% Pass@1 after supervised fine-tuning and 71.7% after reinforcement learning, with additional gains reported on two external benchmarks.

Shared scheduler enables realistic training and testing of continual-learning agents

Jung et al. built an execution substrate for evaluating and post-training agents across sessions, background jobs, memory consolidation and simulated weeks of activity. Across seven ported benchmarks, they found that external memory did not consistently improve on harness-native memory, but post-training a 4-billion-parameter model to use its harness and memory produced substantial efficiency and accuracy gains.

Jointly routing models and harnesses improves agent performance under sparse data

Chen et al. treat agent deployment as a joint routing problem: choosing both the language model and the harness that manages tools, observations, and execution. Across a benchmark spanning 73 models and 25 harnesses, their factorized router outperformed learned baselines and retained useful routing ability when most model-harness combinations lacked training results.

Receiver-specific cache sharing cuts multi-agent costs while improving accuracy

Rossi et al. introduce receiver-conditioned latent communication, in which agents exchange selected internal model states rather than generated summaries or complete key-value caches. On FanOutQA with Qwen 3, their training-free CacheBack method removed 75% of the transferred state while outperforming text communication in both accuracy and latency.

App deeplinks help mobile agents navigate faster and finish more tasks

Sun et al. built a pipeline that extracts app-native deeplinks, tests them on physical devices, and records descriptions of the screens they reach. Their GUI-Hopper agent learns to choose between these direct navigation shortcuts and ordinary GUI actions, improving task success and reducing interaction steps on both benchmark and commercial Android applications.

Regularization limits agent skill growth without sacrificing downstream performance

Nie et al. treat repeated changes to an agent’s reusable instructions as a learning process that can overfit recent experience. Across three skill-evolution benchmarks, their framework constrained unnecessary skill growth, preserved competitive task performance, and detected behavioral regressions missed by structural measures.

Agent certification can fail when proposal and history correlations are erased

Zhong and Zhu model an agent harness that observes a foundation model’s proposed action before deciding whether and how to execute it. They show that replacing history-conditioned proposals with a state-only set can preserve all possible proposals yet eliminate valid safety certificates by introducing combinations of histories and proposals that cannot actually occur together.

LAM Cuts Agent Memory While Bounding Retrieval Score Errors

The authors introduce LAM, a memory framework that deduplicates agent observations without using an LLM summarizer, preserves cached context prefixes, and estimates compaction costs before deployment. It removed 22.47% of observation tokens while retaining 99.984% of measured evidence associated with correct patches, although its mathematical bound covers retrieval-score changes rather than guaranteeing identical retrieval rankings.

Retrieving real states improves long-horizon planning over generated latent sub-goals

Lee et al. identify a failure mode in hierarchical latent planning: macro planners can generate intermediate targets that do not correspond to physically possible states. Their alternative, Metro-WM, searches a graph of previously observed frames and improves long-horizon success by as much as 37.33 percentage points over the next-best hierarchical method.