Scientific research proceeds through iterative cycles of hypothesis generation, experiment design, execution, and revision. AI agents can automate parts of this process, but existing approaches typically follow a single research trajectory or coordinate through a central planner with fixed objectives. As a result, they struggle to sustain parallel exploration, adapt as experimental evidence changes, or preserve knowledge of failed directions over long-running experiments. We introduce AutoScientists, a decentralized team of AI agents for long-running computational scientific experimentation. Agents interpret a shared experimental state, self-organize into teams around promising hypotheses, critique proposals before using experimental compute, and share successes and failures to reduce redundant exploration. Under matched experimental budgets, AutoScientists improves over prior AI agents across biomedical machine learning, language-model training optimization, and protein fitness prediction. On BioML-Bench, spanning biomedical imaging, protein engineering, single-cell omics, and drug discovery, AutoScientists achieves a mean leaderboard percentile of 74.4% across 24 tasks, improving
Long-running LLM agents require persistent memory to preserve state across interactions, yet most deployed systems manage memory with age-based retention (e.g., TTL). While TTL bounds item lifetime, it does not bound the computational footprint of memory on the request path: as retained items accumulate, retrieval candidate sets and vector similarity scans can grow unpredictably, yielding heavy-tailed latency and unstable throughput. We present AMV-L (Adaptive Memory Value Lifecycle), a memory-management framework that treats agent memory as a managed systems resource. AMV-L assigns each memory item a continuously updated utility score and uses value-driven promotion, demotion, and eviction to maintain lifecycle tiers; retrieval is restricted to a bounded, tier-aware candidate set that decouples the request-path working set from total retained memory. We implement AMV-L in a full-stack LLM serving system and evaluate it under identical long-running workloads against two baselines: TTL and an LRU working-set policy, with fixed prompt-injection caps. Relative to TTL, AMV-L improves throughput by 3.1x and reduces latency by 4.2x (median), 4.7x (p95), and 4.4x (p99), while reducing the
Long-running AI agents fail not only when inference fails or tools are underspecified, but when independently evolving model and harness layers change the semantics of belief, capability, and goal commitments across their boundary - a failure class this paper terms Interface Volatility. This paper argues that Agent Epistemic Integrity (AEI) must be treated as a first-class architectural constraint, achievable only through joint model-harness design organized around an explicit interface contract. The central claim is that the model-harness interface contract is the precondition for joint design; its operational form is a four-level hierarchy - goal validity, action-archetype sequencing, tool-instance selection, and invocation-level failure discrimination - that specifies what the boundary must preserve and what structured outputs the model must return for the contract to hold across levels. This reframes long-running agent design away from flat action loops and toward contract-preserving control over persistent state. Evaluation and training should therefore derive from the contract itself, testing whether belief, tool, and goal commitments hold across session boundaries and indepe
Long-running autonomous AI agents suffer from a well-documented memory coherence problem: tool-execution success rates degrade 14 percentage points over 72-hour operation windows due to four compounding failure modes in existing flat-file memory systems. We present MEMTIER, a tripartite memory architecture for the OpenClaw agent runtime that introduces a structured episodic JSONL store, a five-signal weighted retrieval engine, an attention-attributed cognitive weight update loop, an asynchronous consolidation daemon promoting episodic facts to a semantic tier, and a PPO-based policy framework for adapting retrieval weights (infrastructure validated; performance gains pending camera-ready). On the full 500-question LongMemEval-S benchmark (Wu et al., 2025), MEMTIER achieves Acc=0.382, F1=0.412 with Qwen2.5-7B on a consumer 6GB GPU - a +33 percentage point improvement over the full-context baseline (0.050 -> 0.382, i.e., 5% -> 38%). With DeepSeek-V4-Flash fact pre-population, single-session recall reaches 0.686-0.714, exceeding the paper's RAG BM25 GPT-4o baseline (0.560) on those categories. Temporal reasoning rises to 0.323 and multi-session synthesis to 0.173, demonstrating
AI agents are increasingly asked to carry out work that spans minutes, hours, or longer. Yet the default model of agent behavior is continuous action: issuing tool calls, refreshing pages, searching for alternatives, or otherwise trying to force progress. This is the wrong approach for many long-running tasks, which are better served by a strategy of sustained attention. Instead, agents should monitor an environment, notice when an external event makes progress possible, then respond promptly without wasting resources while waiting. To measure progress on this class of tasks, we introduce SentinelBench, an open-source benchmark for time-evolving monitoring tasks. SentinelBench contains 100 tasks across 10 synthetic web environments, including email, calendars, finance, professional networking, and entertainment. Each environment exposes a live web interface and replays a scripted sequence of events, requiring agents to navigate and reason about web pages whose state shifts underfoot. SentinelBench measures task completion, reaction time, and resource use, exposing the tradeoff between responsiveness and cost. We report results across three models and two browser-agent harnesses, es
Large Language Models (LLMs) often experience performance degradation during long-running interactions due to increasing context length, memory saturation, and computational overhead. This paper presents an adaptive context compression framework that integrates importance-aware memory selection, coherence-sensitive filtering, and dynamic budget allocation to retain essential conversational information while controlling context growth. The approach is evaluated on LOCOMO, LOCCO, and LongBench benchmarks to assess answer quality, retrieval accuracy, coherence preservation, and efficiency. Experimental results demonstrate that the proposed method achieves consistent improvements in conversational stability and retrieval performance while reducing token usage and inference latency compared with existing memory and compression-based approaches. These findings indicate that adaptive context compression provides an effective balance between long-term memory preservation and computational efficiency in persistent LLM interactions
Long-running language agents need more than memory access. Retrieval systems can fetch past facts at query time, but they do not decide which experiences should continue to shape behavior after the working context is unloaded. We study this separate problem as memory depth: durable goal-conditioned tendencies written into a small parametric store. We introduce the loop-drift protocol, a controlled stress test in which the retrieval index remains intact while working context is unloaded and goal-conditioned behavior must persist under long-loop interference. We evaluate EVAF, a surprise- and valence-gated LoRA consolidation mechanism. Across GPT-2 and TinyLlama, retrieval is strongest on shallow factual recall (short-fact accuracy 0.956--0.973), while EVAF is strongest on goal persistence and post-unload recovery (0.812--0.904) with only 2--3 parametric writes per 200 events. Mechanism controls show that selective consolidation factorizes into two controllable dimensions: selection and actuation. Matched random gates isolate selection beyond sparse writing; fixed-inner controls across GPT-2, TinyLlama, and Mistral-7B show that inner-loop write strength is model-dependent; and a Mist
Memory systems often organize user-agent interactions as retrievable external memory and are crucial for long-running agents by overcoming the limited context windows of LLMs. However, existing memory systems invoke LLMs to process every incoming interaction for memory extraction, and such an eager memory consolidation scheme leads to substantial token consumption. To tackle this problem, we propose RecMem by rethinking when memory consolidation should be conducted. RecMem stores incoming interactions in a subconscious memory layer and encode them using lightweight embedding models for retrieval. LLMs are only invoked to extract episodic and semantic memory when sustained recurrence are observed for semantically similar interactions. Such recurrence-based consolidation works because these interactions correspond to a semantic cluster with rich information and thus are worth extraction and summarization. To improve accuracy, RecMem also incorporates a semantic refinement mechanism that recovers the fine-grained facts omitted by memory extraction. Experiments show that RecMem reduces the memory construction token cost of three SOTA memory systems by up to 87% while exceeding their ac
Modern AI agents increasingly combine conversational interaction with autonomous task execution, such as coding and web research, raising a natural question: What happens when an agent engaged in long-horizon tasks is exposed to user persuasion? Yet studying this possibility is challenging because long-running agent behavior is noisy and costly to reproduce, and it remains unclear which unique challenges emerge only in extended task execution. We study how belief-level intervention can influence downstream task behavior, a phenomenon we name persuasion propagation. We introduce a behavior-centered evaluation framework that distinguishes between persuasion applied during or prior to task execution. Across web research and coding tasks, we find that on-the-fly persuasion induces weak and inconsistent behavioral effects. In contrast, when the belief state is explicitly specified at task time, belief-prefilled agents conduct on average 26.9% fewer searches and visit 16.9% fewer unique sources than neutral-prefilled agents. These results suggest that persuasion, even in prior interaction, can affect the agent's behavior, motivating behavior-level evaluation in agentic systems.
The rise of AI-native Low-Code/No-Code (LCNC) platforms enables autonomous agents capable of executing complex, long-duration business processes. However, a fundamental challenge remains: memory management. As agents operate over extended periods, they face "memory inflation" and "contextual degradation" issues, leading to inconsistent behavior, error accumulation, and increased computational cost. This paper proposes a novel hybrid memory system designed specifically for LCNC agents. Inspired by cognitive science, our architecture combines episodic and semantic memory components with a proactive "Intelligent Decay" mechanism. This mechanism intelligently prunes or consolidates memories based on a composite score factoring in recency, relevance, and user-specified utility. A key innovation is a user-centric visualization interface, aligned with the LCNC paradigm, which allows non-technical users to manage the agent's memory directly, for instance, by visually tagging which facts should be retained or forgotten. Through simulated long-running task experiments, we demonstrate that our system significantly outperforms traditional approaches like sliding windows and basic RAG, yielding
This paper proposes Oze, a concurrency control protocol that handles heterogeneous workloads, including long-running update transactions. Oze explores a large scheduling space using a multi-version serialization graph to reduce false positives. Oze manages the graph in a decentralized manner to exploit many cores in modern servers. We further propose an OLTP benchmark, BoMB (Bill of Materials Benchmark), based on a use case in an actual manufacturing company. BoMB consists of one long-running update transaction and five short transactions that conflict with each other. Experiments using BoMB show that Oze can handle the long-running update transaction while achieving four orders of magnitude higher throughput than state-of-the-art optimistic and multi-version protocols and up to five times higher throughput than pessimistic protocols. We also show Oze performs comparably with existing techniques even in a typical OLTP workload, TPC-C, thanks to a protocol switching mechanism.
Spot instances offer a cost-effective solution for applications running in the cloud computing environment. However, it is challenging to run long-running jobs on spot instances because they are subject to unpredictable evictions. Here, we present Spot-on, a generic software framework that supports fault-tolerant long-running workloads on spot instances through checkpoint and restart. Spot-on leverages existing checkpointing packages and is compatible with the major cloud vendors. Using a genomics application as a test case, we demonstrated that Spot-on supports both application-specific and transparent checkpointing methods. Compared to running applications using on-demand instances, it allows the completion of these workloads for a significant reduction in computing costs. Compared to running applications using application-specific checkpoint mechanisms, transparent checkpoint-protected applications reduce runtime by up to 40%, leading to further cost savings of up to 86%.
Resource provisioning plays a pivotal role in determining the right amount of infrastructure resource to run applications and target the global decarbonization goal. A significant portion of production clusters is now dedicated to long-running applications (LRAs), which are typically in the form of microservices and executed in the order of hours or even months. It is therefore practically important to plan ahead the placement of LRAs in a shared cluster so that the number of compute nodes required by them can be minimized to reduce carbon footprint and lower operational costs. Existing works on LRA scheduling are often application-agnostic, without particularly addressing the constraining requirements imposed by LRAs, such as co-location affinity constraints and time-varying resource requirements. In this paper, we present an affinity-aware resource provisioning approach for deploying large-scale LRAs in a shared cluster subject to multiple constraints, with the objective of minimizing the number of compute nodes in use. We investigate a broad range of solution algorithms which fall into three main categories: Application-Centric, Node-Centric, and Multi-Node approaches, and tune
Reliability, longevity, availability, and deadline guarantees are the four most important metrics to measure the QoS of long-running safety-critical real-time applications. Software aging is one of the major factors that impact the safety of long-running real-time applications as the degraded performance and increased failure rate caused by software aging can lead to deadline missing and catastrophic consequences. Software rejuvenation is one of the most commonly used approaches to handle issues caused by software aging. In this paper, we study the optimal time when software rejuvenation shall take place so that the system's reliability, longevity, and availability are maximized, and application delays caused by software rejuvenation is minimized. In particular, we formally analyze the relationships between software rejuvenation frequency and system reliability, longevity, and availability. Based on the theoretic analysis, we develop approaches to maximizing system reliability, longevity, and availability, and use simulation to evaluate the developed approaches. In addition, we design the MIN-DELAY semi-priority-driven scheduling algorithm to minimize application delays caused by r
This paper presents Tartarian, a tool that supports maintenance of software with long-running, multi-release branches in distributed version control systems. When new maintenance code, such as bug fixes and code improvement, is committed into a branch, it is likely that such code can be applied or reused with some other branches. To do so, a developer may manually identify a commit and cherry pick it. Tartarian can support this activity by providing commit hashtags, which the developer uses as metadata to specify their intentions when committing the code. With these tags, Tartarian uses dependency graph, that represents the dependency constraints of the branches, and Branch Identifier, which matches the commit hashtags with the dependency graph, to identify the applicable branches for the commits. Using Tartarian, developers may be able to maintain software with multiple releases more efficiently.
In this paper, we describe a novel proactive recovery scheme based on service migration for long-running Byzantine fault tolerant systems. Proactive recovery is an essential method for ensuring long term reliability of fault tolerant systems that are under continuous threats from malicious adversaries. The primary benefit of our proactive recovery scheme is a reduced vulnerability window. This is achieved by removing the time-consuming reboot step from the critical path of proactive recovery. Our migration-based proactive recovery is coordinated among the replicas, therefore, it can automatically adjust to different system loads and avoid the problem of excessive concurrent proactive recoveries that may occur in previous work with fixed watchdog timeouts. Moreover, the fast proactive recovery also significantly improves the system availability in the presence of faults.
Improving social welfare is a complex challenge requiring policymakers to optimize objectives across multiple time horizons. Evaluating the impact of such policies presents a fundamental challenge, as those that appear suboptimal in the short run may yield significant long-term benefits. We tackle this challenge by analyzing the long-term dynamics of two prominent policy frameworks: Rawlsian policies, which prioritize those with the greatest need, and utilitarian policies, which maximize immediate welfare gains. Conventional wisdom suggests these policies are at odds, as Rawlsian policies are assumed to come at the cost of reducing the average social welfare, which their utilitarian counterparts directly optimize. We challenge this assumption by analyzing these policies in a sequential decision-making framework where individuals' welfare levels stochastically decay over time, and policymakers can intervene to prevent this decay. Under reasonable assumptions, we prove that interventions following Rawlsian policies can outperform utilitarian policies in the long run, even when the latter dominate in the short run. We characterize the exact conditions under which Rawlsian policies can
The literature on panel cointegration is extensive but does not cover data sets where the cross section dimension, $n$, is larger than the time series dimension $T$. This paper proposes a novel methodology that filters out the short run dynamics using sub-sample time averages as deviations from their full-sample counterpart, and estimates the number of long-run relations and their coefficients using eigenvalues and eigenvectors of the pooled covariance matrix of these sub-sample deviations. We refer to this procedure as pooled minimum eigenvalue (PME). We show that PME estimator is consistent and asymptotically normal as $n$ and $T \rightarrow \infty$ jointly, such that $T\approx n^{d}$, with $d>0$ for consistency and $d>1/2$ for asymptotic normality. Extensive Monte Carlo studies show that the number of long-run relations can be estimated with high precision, and the PME estimators have good size and power properties. The utility of our approach is illustrated by micro and macro applications using Compustat and Penn World Tables.
We study a long-run persuasion problem where a long-lived Sender repeatedly interacts with a sequence of short-lived Receivers who may adopt a misspecified model for belief updating. The Sender commits to a stationary information structure, but suspicious Receivers compare it to an uninformative alternative and may switch based on the Bayes factor rule. We characterize when the one-shot Bayesian Persuasion-optimal (BP-optimal) structure remains optimal in the long run despite this switching risk. In particular, when Receivers cannot infer the state from the Sender's preferred action, they never switch, and the BP-optimal structure maximizes the Sender's lifetime utility. In contrast, when such inference is possible, full disclosure may outperform BP-optimal. Our findings highlight the strategic challenges of information design when the Receivers' interpretation of signals evolves over time.
In the paper we study continuous time controlled Markov processes using discrete time controlled Markov processes. We consider long run functionals: average reward per unit time or long run risk sensitive functional. We also investigate stability of continuous time functionals with respect to pointwise convergence of Markov controls.