<?xml version="1.0" encoding="utf-8"?>
<!DOCTYPE article PUBLIC "-//NLM//DTD JATS (Z39.96) Journal Publishing DTD v1.0 20120330//EN" "http://jats.nlm.nih.gov/publishing/1.0/JATS-journalpublishing1.dtd">
<article xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" dtd-version="1.0" article-type="other">
  <front>
    <journal-meta>
      <journal-id journal-id-type="nlm-ta">AI Agent</journal-id>
      <journal-id journal-id-type="publisher-id">aiagent</journal-id>
      <journal-title-group>
        <journal-title>AI Agent</journal-title>
      </journal-title-group>
      <issn pub-type="epub">3070-3719</issn>
      <publisher>
        <publisher-name>OAE Publishing Inc.</publisher-name>
      </publisher>
    </journal-meta>
    <article-meta>
      <article-id pub-id-type="doi">10.20517/aiagent.2026.37</article-id>
      <article-id pub-id-type="publisher-id">AIAgent-2026-37</article-id>
      <article-categories>
        <subj-group>
          <subject>Perspective</subject>
        </subj-group>
      </article-categories>
      <title-group>
        <article-title>AI agents for science and society in 2025: ten representative advances</article-title>
      </title-group>
      <contrib-group>
        <contrib contrib-type="author" corresp="yes">
          <name>
            <surname>Li</surname>
            <given-names>Limin</given-names>
          </name>
          <xref ref-type="aff" rid="I1">
            <sup>1</sup>
          </xref>
          <xref ref-type="aff" rid="I2">
            <sup>2</sup>
          </xref>
          <xref ref-type="corresp" rid="cor1">*</xref>
        </contrib>
      </contrib-group>
      <aff id="I1"><sup>1</sup>Suzhou MatSource Technology Co., Ltd., Suzhou 215000, Jiangsu, China.</aff>
      <aff id="I2"><sup>2</sup>Gusu Laboratory of Materials, Suzhou 215000, Jiangsu, China.</aff>
      <author-notes>
        <corresp id="cor1">Correspondence to: Dr. Limin Li, Suzhou MatSource Technology Co., Ltd., Suzhou 215000, Jiangsu, China. E-mail: <email>hotline@matsourceai.com</email></corresp>
        <fn fn-type="other">
          <p><bold>Received:</bold> 29 Jun 2026 | <bold>First Decision:</bold> 22 Jul 2026 | <bold>Revised:</bold> 14 Oct 2026 | <bold>Accepted:</bold> 22 Oct 2026 | <bold>Published:</bold> 18 Nov 2026</p>
        </fn>
        <fn fn-type="other">
          <p><bold>Academic Editor:</bold> Jianjun Liu | <bold>Copy Editor:</bold> Tong Wang | <bold>Production Editor:</bold> Tong Wang</p>
        </fn>
      </author-notes>
      <pub-date pub-type="ppub">
        <year>2026</year>
      </pub-date>
      <pub-date pub-type="epub">
        <day>18</day>
        <month>11</month>
        <year>2026</year>
      </pub-date>
      <volume>2</volume>
	  <issue>4</issue>
      <elocation-id>34</elocation-id>
      <permissions>
        <copyright-statement>© The Author(s) 2026.</copyright-statement>
        <license xlink:href="https://creativecommons.org/licenses/by/4.0/">
          <license-p>© The Author(s) 2026.<bold>Open Access</bold>This article is licensed under a Creative Commons Attribution 4.0 International License (<uri xlink:href="https://creativecommons.org/licenses/by/4.0/">https://creativecommons.org/licenses/by/4.0/</uri>), which permits unrestricted use, sharing, adaptation, distribution and reproduction in any medium or format, for any purpose, even commercially, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.</license-p>
        </license>
      </permissions>
    </article-meta>
  </front>
  <body>
    <p>Artificial intelligence (AI) agents are transitioning from conversational add-ons to integrated scientific operating layers, reflecting a broader shift toward autonomous systems capable of reasoning, planning, and interacting with external tools<sup>[<xref ref-type="bibr" rid="B1">1</xref>]</sup>. Advances around 2025 instead embed models in structured loops for planning, tool use, feedback, and revision. This Perspective highlights ten representative advances spanning general agent architectures, mathematical reasoning, chemistry, materials science, biomedicine, scientific instrumentation, energy systems, Earth observation, embodied intelligence, and security governance. Collectively, these developments mark a transition from fluent text generation toward controllable, auditable, and increasingly autonomous cycles of reasoning and action.</p>
    <p>To identify representative advances in AI agents for science and society, we considered four criteria: scientific significance, agentic functionality, cross-domain representation, and demonstrated progress and influence during 2024-2025. Particular emphasis was placed on systems that move beyond passive information generation by combining planning, tool utilization, feedback-driven adaptation, and iterative decision-making with scientific data, computational resources, physical environments, and human expertise. The selected advances represent major developmental directions rather than a ranked list of individual studies.</p>
    <sec id="sec1">
      <title>FOUNDATIONS AND GENERAL ARCHITECTURES OF AI AGENTS</title>
      <sec id="sec1-1">
        <title>Modular planning and long-horizon agent architectures</title>
        <p>Long-horizon agency exposes a fundamental limitation of monolithic prompting: a single stream of text must simultaneously retain goals, track state, follow constraints, and anticipate long-term consequences. Recent agent architectures have therefore explored explicit reasoning traces, external memory, and iterative interaction mechanisms to improve long-horizon execution<sup>[<xref ref-type="bibr" rid="B2">2</xref>-<xref ref-type="bibr" rid="B4">4</xref>]</sup>. Errors may accumulate over extended tasks, motivating inspectable architectures<sup>[<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B5">5</xref>]</sup>. <xref ref-type="fig" rid="fig1">Figure 1</xref> summarizes this emerging design pattern.</p>
        <fig id="fig1" position="float">
          <label>Figure 1</label>
          <caption>
            <p>Modular planning architecture linking goal decomposition, action proposal, constraint checking, state prediction, outcome evaluation, and feedback.</p>
          </caption>
          <graphic xlink:href="aiagent2037.fig.1.jpg"/>
        </fig>
        <p>Rather than relying on an opaque prompt, modular systems separate functions such as goal decomposition, action proposal, feasibility monitoring, future-state prediction, outcome evaluation, and plan coordination<sup>[<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B7">7</xref>]</sup>. Feedback can be routed back to earlier stages, turning planning into a controlled search process and making intermediate assumptions visible<sup>[<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B8">8</xref>]</sup>.</p>
        <p>The main significance of this architecture lies in its debuggability and extensibility: individual modules can be tested, replaced, or constrained independently, while the agent can be coupled to simulators, specialized tools, and long-term memory<sup>[<xref ref-type="bibr" rid="B6">6</xref>-<xref ref-type="bibr" rid="B8">8</xref>]</sup>. Robust coordination under noisy observations, non-stationary environments, and limited computational resources remains an important challenge<sup>[<xref ref-type="bibr" rid="B5">5</xref>-<xref ref-type="bibr" rid="B8">8</xref>]</sup>.</p>
      </sec>
    </sec>
    <sec id="sec2">
      <title>MATHEMATICAL REASONING AND FORMAL METHODS</title>
      <sec id="sec2-1">
        <title>Verifiable reasoning through formal methods and search</title>
        <p>Mathematical agents require verification protocols that go beyond semantic plausibility. Because a single-step error can invalidate a long logical chain, recent research has coupled large language models (LLMs) with proof assistants, formula checkers, and explicit search algorithms to enforce correctness through deterministic environments rather than statistical plausibility alone<sup>[<xref ref-type="bibr" rid="B9">9</xref>-<xref ref-type="bibr" rid="B11">11</xref>]</sup>. The verification loop is summarized in <xref ref-type="fig" rid="fig2">Figure 2</xref>.</p>
        <fig id="fig2" position="float">
          <label>Figure 2</label>
          <caption>
            <p>Formal reasoning loop combining problem formalization, tactic selection, verification feedback, search, and policy improvement.</p>
          </caption>
          <graphic xlink:href="aiagent2037.fig.2.jpg"/>
        </fig>
        <p>A typical workflow formalizes a natural-language problem into a symbolic representation, proposes proof tactics, evaluates them within a verification engine, and expands the search over valid proof states. Checker feedback identifies progress, failure, or proof completion. These trajectories can subsequently improve search policies and iterative reasoning.</p>
        <p>This framework shifts agentic reasoning toward auditable computation and is relevant to theorem proving, program verification, symbolic derivation, and other scientific workflows governed by hard constraints. Remaining bottlenecks include the cost of formalization, expanding search spaces, and the interpretability gap between machine-checkable proofs and human mathematical intuition.</p>
      </sec>
    </sec>
    <sec id="sec3">
      <title>CHEMICAL SCIENCES AND AUTONOMOUS DISCOVERY</title>
      <sec id="sec3-1">
        <title>Multi-agent orchestration and closed-loop chemical research</title>
        <p>Autonomous chemistry requires coordination across diverse representations and infrastructures, including literature, reaction logic, experimental protocols, instruments, measurements, and predictive models. Recent methodologies have therefore moved beyond single-agent chatbots toward multi-agent and tool-integrated workflows. <xref ref-type="fig" rid="fig3">Figure 3</xref> illustrates how literature understanding, computation, robotic execution, and data analysis can be connected within a closed-loop workflow<sup>[<xref ref-type="bibr" rid="B12">12</xref>-<xref ref-type="bibr" rid="B14">14</xref>]</sup>.</p>
        <fig id="fig3" position="float">
          <label>Figure 3</label>
          <caption>
            <p>Multi-agent chemistry workflow connecting research planning, computational analysis, laboratory operations, and experimental feedback.</p>
          </caption>
          <graphic xlink:href="aiagent2037.fig.3.jpg"/>
        </fig>
        <p>Specialist agents can perform localized tasks: one may extract prior knowledge from literature, another generate hypotheses or synthetic protocols, while computational or robotic modules carry out prediction, synthesis, characterization, or analysis. DigCat provides catalyst data and AI-enabled design resources<sup>[<xref ref-type="bibr" rid="B15">15</xref>]</sup>, Cloud Synthesis describes catalyst design with synthesis and feedback<sup>[<xref ref-type="bibr" rid="B16">16</xref>]</sup>, and DIVE extracts literature-figure data for hydrogen-storage materials discovery<sup>[<xref ref-type="bibr" rid="B17">17</xref>]</sup>. Robotic laboratory studies demonstrate experimental execution and feedback<sup>[<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B14">14</xref>]</sup>. Practical challenges remain in safe instrument control, exception handling, protocol transferability, data quality assurance, and reproducibility.</p>
      </sec>
    </sec>
    <sec id="sec4">
      <title>MATERIALS SCIENCE AND INTELLIGENT DESIGN</title>
      <sec id="sec4-1">
        <title>Physics-grounded agents for materials design and simulation</title>
        <p>Materials discovery is inherently a high-dimensional, multimodal reasoning problem. Viable candidates must satisfy criteria including thermodynamic stability, kinetic accessibility, structural symmetry, synthetic feasibility, and target performance. General-purpose LLMs still show limitations in domain-specific physical reasoning, quantitative reliability, and robust materials-property prediction, motivating their integration with curated materials data and physics-based computational tools<sup>[<xref ref-type="bibr" rid="B18">18</xref>]</sup>. <xref ref-type="fig" rid="fig4">Figure 4</xref> summarizes a representative physics-aware materials-agent architecture.</p>
        <fig id="fig4" position="float">
          <label>Figure 4</label>
          <caption>
            <p>Physics-aware materials-agent architecture connecting retrieval, shared memory, planning, computation, simulation, and analysis.</p>
          </caption>
          <graphic xlink:href="aiagent2037.fig.4.jpg"/>
        </fig>
        <p>DIVE extracts materials data from scientific figures<sup>[<xref ref-type="bibr" rid="B17">17</xref>]</sup>, while Merchant <italic>et al.</italic> demonstrate large-scale learning for materials stability prediction<sup>[<xref ref-type="bibr" rid="B19">19</xref>]</sup>. Materials databases, including the Materials Project<sup>[<xref ref-type="bibr" rid="B20">20</xref>]</sup>, and catalytic data resources<sup>[<xref ref-type="bibr" rid="B15">15</xref>]</sup> can support screening and property prediction.</p>
        <p>A shared memory or coordination layer can connect knowledge retrieval, inverse design, code execution, simulation, and result analysis, allowing the agent to function as an orchestration layer rather than an isolated predictor. Cloud Synthesis links catalyst design to experimental feedback<sup>[<xref ref-type="bibr" rid="B16">16</xref>]</sup>, while DIVE connects materials data extraction to inverse design<sup>[<xref ref-type="bibr" rid="B17">17</xref>]</sup>. Important challenges remain in uncertainty quantification, reliable tool interfaces, efficient simulations, and integration with experimental validation.</p>
      </sec>
    </sec>
    <sec id="sec5">
      <title>BIOMEDICINE AND THERAPEUTIC INNOVATION</title>
      <sec id="sec5-1">
        <title>Tool-constrained biomedical experimentation</title>
        <p>Biomedical design operates under strict biological, ethical, and safety constraints, where parameter changes can affect pharmacokinetics, target selectivity, and experimental feasibility. Biomedical foundation models have advanced protein structure prediction<sup>[<xref ref-type="bibr" rid="B21">21</xref>-<xref ref-type="bibr" rid="B23">23</xref>]</sup>. <xref ref-type="fig" rid="fig5">Figure 5</xref> summarizes a tool-constrained framework in which biological objectives are translated into structured plans and computational tasks.</p>
        <fig id="fig5" position="float">
          <label>Figure 5</label>
          <caption>
            <p>Biomedical-agent framework linking biological objectives, planning, specialized computational tools, feedback, and human oversight. QC: Quality control; NPs: nanoparticles.</p>
          </caption>
          <graphic xlink:href="aiagent2037.fig.5.jpg"/>
        </fig>
        <p>The agent can decompose an objective into target selection, sequence or structure design, protocol matching, evaluation, and failure-mode prediction. Biomni integrates biomedical reasoning with tools and code execution<sup>[<xref ref-type="bibr" rid="B24">24</xref>]</sup>, while BioDiscoveryAgent proposes genetic perturbation experiments<sup>[<xref ref-type="bibr" rid="B25">25</xref>]</sup>.</p>
        <p>This paradigm reflects a broader shift from standalone language models toward biomedical agents that coordinate domain-specific tools under explicit experimental constraints. Its practical value depends on biological validation, human supervision, laboratory safety, and ethical review.</p>
      </sec>
    </sec>
    <sec id="sec6">
      <title>PHYSICS, ASTRONOMY, AND SPACE EXPLORATION</title>
      <sec id="sec6-1">
        <title>Natural-language scientific instrument control</title>
        <p>Modern scientific instruments are increasingly powerful but often rely on specialized, manually intensive control interfaces. Natural language can lower this barrier only if user intent is translated faithfully into structured, safe, and machine-readable commands. Recent work has explored LLM-based agents as semantic interfaces between scientists and instrument-control software, with explicit validation and feedback<sup>[<xref ref-type="bibr" rid="B26">26</xref>,<xref ref-type="bibr" rid="B27">27</xref>]</sup>. The control loop shown in <xref ref-type="fig" rid="fig6">Figure 6</xref> illustrates this interaction.</p>
        <fig id="fig6" position="float">
          <label>Figure 6</label>
          <caption>
            <p>Workflow of scientific instrument control, from user-intent interpretation and structured prompt formulation to machine-readable command generation, instrument execution, observation, and feedback. LLM: Large language model; BPM: beam position monitor.</p>
          </caption>
          <graphic xlink:href="aiagent2037.fig.6.jpg"/>
        </fig>
        <p>A representative workflow parses user intent into goals and constraints, generates executable commands, and uses returned measurements or device status to guide subsequent adjustments. The LLM therefore serves as a reasoning and translation layer rather than an unconstrained controller.</p>
        <p>The same principle can extend to accelerators, microscopes, beamlines, telescopes, and robotic measurement platforms. Reliable deployment requires safety interlocks, operator authority, validated command schemas, and deterministic fallback mechanisms<sup>[<xref ref-type="bibr" rid="B26">26</xref>,<xref ref-type="bibr" rid="B27">27</xref>]</sup>.</p>
      </sec>
    </sec>
    <sec id="sec7">
      <title>ENERGY SYSTEMS AND DECARBONIZATION</title>
      <sec id="sec7-1">
        <title>Autonomous data-to-algorithm loops for energy systems</title>
        <p>Modern energy systems must adapt continuously to variable generation and demand while converting operational data into reliable control and forecasting decisions. Data-driven methods have shown value for prediction, optimization, and adaptive energy management<sup>[<xref ref-type="bibr" rid="B28">28</xref>]</sup>. <xref ref-type="fig" rid="fig7">Figure 7</xref> outlines an iterative workflow linking task formulation, system data, optimization, evaluation, and feedback.</p>
        <fig id="fig7" position="float">
          <label>Figure 7</label>
          <caption>
            <p>Energy-systems workflow linking task formulation, data collection, iterative search, evaluation, and feedback. ML: Machine learning; APIs: application programming interfaces; SCADA: supervisory control and data acquisition.</p>
          </caption>
          <graphic xlink:href="aiagent2037.fig.7.jpg"/>
        </fig>
        <p>Within this loop, incoming data support state assessment, optimization-task formulation, and evaluation of candidate control or forecasting strategies. Reinforcement-learning and autonomous-optimization methods provide a foundation for adaptive behavior, where policies improve through interaction with changing environments<sup>[<xref ref-type="bibr" rid="B29">29</xref>]</sup>.</p>
        <p>This approach is relevant to grid optimization, dynamic state estimation, integrated energy management, and resilient operation. Translation to real systems depends on reliable data pipelines, interpretable failures, operational safeguards, and clear accountability under uncertainty.</p>
      </sec>
    </sec>
    <sec id="sec8">
      <title>EARTH AND ENVIRONMENTAL INTELLIGENCE</title>
      <sec id="sec8-1">
        <title>Tool-augmented multimodal reasoning for Earth observation</title>
        <p>Earth observation (EO) is inherently multimodal, combining optical, multispectral or hyperspectral imagery, SAR, elevation models, land-cover products, climate data, and contextual information. Complex EO tasks therefore benefit from tool-augmented reasoning rather than single-image classification alone<sup>[<xref ref-type="bibr" rid="B30">30</xref>,<xref ref-type="bibr" rid="B31">31</xref>]</sup>. <xref ref-type="fig" rid="fig8">Figure 8</xref> presents a workflow coupling multimodal inputs with geospatial tools and iterative reasoning.</p>
        <fig id="fig8" position="float">
          <label>Figure 8</label>
          <caption>
            <p>Earth-observation agent combining multimodal imagery, geospatial tools, iterative reasoning, and quantitative evaluation. EO: Earth observation; RGB: red, green, and blue; SAR: synthetic aperture radar; DEM: digital elevation model; POI: point of interest; ML: machine learning; AI: artificial intelligence; EM: exact match; F1: F1 score (harmonic mean of precision and recall); mIoU: mean intersection over union; OOD: out of distribution.</p>
          </caption>
          <graphic xlink:href="aiagent2037.fig.8.jpg"/>
        </fig>
        <p>A generic EO agent follows a reason-act-observe cycle: it interprets the task, selects tools, processes intermediate outputs, and updates its state until an interpretation or quantitative result is supported by evidence. Tools include spectral analysis, geometry, retrieval, visualization, and AI models.</p>
        <p>This approach brings agentic AI closer to practical geospatial analysis, where conclusions depend on layered and measurable evidence. Open challenges include sensor shift, missing metadata, uncertainty quantification, and reliable evaluation of tool use and reasoning trajectories<sup>[<xref ref-type="bibr" rid="B30">30</xref>]</sup>.</p>
      </sec>
    </sec>
    <sec id="sec9">
      <title>EMBODIED INTELLIGENCE AND AUTONOMOUS SYSTEMS</title>
      <sec id="sec9-1">
        <title>Feedback-driven embodied intelligence</title>
        <p>Embodied agents operate under contact mechanics, non-equilibrium conditions, and sensorimotor uncertainty, requiring tight coupling between high-level planning and low-level feedback. Vision-language-action models have demonstrated the potential to combine language-based reasoning with multimodal perception and robotic control, translating abstract instructions into executable physical behaviors<sup>[<xref ref-type="bibr" rid="B32">32</xref>-<xref ref-type="bibr" rid="B34">34</xref>]</sup>. <xref ref-type="fig" rid="fig9">Figure 9</xref> highlights this coupling between planning, execution, and feedback.</p>
        <fig id="fig9" position="float">
          <label>Figure 9</label>
          <caption>
            <p>Embodied-agent workflow connecting language-model planning, task memory, robot execution, and visual and force feedback. LLM: Large language model.</p>
          </caption>
          <graphic xlink:href="aiagent2037.fig.9.jpg"/>
        </fig>
        <p>In a representative pipeline, the system retrieves relevant skills, generates a high-level plan, and passes it to the trajectory planner and robot controller. Visual observations, position estimates, and force feedback return to the controller for online replanning and motion correction.</p>
        <p>Such feedback-driven architectures are better suited to long-horizon manipulation than a single offline plan. Barriers remain in control latency, transfer across platforms, safe human-robot interaction, semantic grounding, and unified evaluation for open-ended physical tasks.</p>
      </sec>
    </sec>
    <sec id="sec10">
      <title>TRUST, SAFETY, AND ETHICAL GOVERNANCE</title>
      <sec id="sec10-1">
        <title>Full-stack security and governance for AI agents</title>
        <p>Agents interacting with tools and physical systems face security risks beyond text generation. Vulnerabilities may arise across prompts, memory, planning, actions, tools, observations, and external environments, as summarized in <xref ref-type="fig" rid="fig10">Figure 10</xref>. Indirect prompt injection can compromise tool-integrated applications<sup>[<xref ref-type="bibr" rid="B35">35</xref>]</sup>. Language-model red-teaming supports adversarial evaluation<sup>[<xref ref-type="bibr" rid="B36">36</xref>]</sup>, while Agent Security Bench assesses agent-specific attacks, including memory poisoning, and defenses<sup>[<xref ref-type="bibr" rid="B37">37</xref>]</sup>.</p>
        <fig id="fig10" position="float">
          <label>Figure 10</label>
          <caption>
            <p>AI agent framework with four attack surfaces and corresponding defense governance. AI: Artificial intelligence; LLM: large language model.</p>
          </caption>
          <graphic xlink:href="aiagent2037.fig.10.jpg"/>
        </fig>
        <p>This full-stack perspective shifts governance from post hoc content moderation toward engineering controls over interaction pathways. The central challenge is maintaining reliable evaluation as agent architectures and deployment environments evolve while balancing autonomy, usefulness, transparency, and security.</p>
      </sec>
    </sec>
    <sec id="sec11">
      <title>CONCLUDING PERSPECTIVES</title>
      <p>Taken together, the ten representative advances reveal a broader transition from systems that primarily generate information toward agents that participate directly in scientific and societal workflows.</p>
      <p>Across these domains, successful agents combine long-horizon planning, domain-specific tools, multimodal information processing, memory and state tracking, and iterative feedback<sup>[<xref ref-type="bibr" rid="B1">1</xref>]</sup>. A particularly important feature is the emergence of closed-loop workflows in which hypothesis generation, tool execution, outcome evaluation, and strategy refinement are connected through interaction with external knowledge, computational models, experimental platforms, or physical environments<sup>[<xref ref-type="bibr" rid="B16">16</xref>,<xref ref-type="bibr" rid="B17">17</xref>]</sup>.</p>
      <p>Several challenges recur across application areas. Long-horizon autonomy remains vulnerable to accumulated planning and tool-use errors, while generalization, uncertainty quantification, standardized evaluation, and reproducibility remain unresolved. Transparency, human oversight, safety constraints, and rigorous validation therefore remain essential for high-impact applications<sup>[<xref ref-type="bibr" rid="B35">35</xref>-<xref ref-type="bibr" rid="B37">37</xref>]</sup>.</p>
      <p>Future development is likely to focus on more reliable integration of foundation models, specialized knowledge, scientific tools, and autonomous experimentation. Rather than replacing human expertise, next-generation agents are more plausibly viewed as collaborative systems that extend scientific reasoning and decision-making while preserving meaningful human supervision and accountability.</p>
    </sec>
  </body>
  <back>
    <sec>
      <title>DECLARATIONS</title>
      <sec>
        <title>Acknowledgments</title>
        <p>The author thanks Hanlin Zhuang from Suzhou MatSource Technology Co., Ltd. for helpful comments on the manuscript and technical assistance with the organization of literature resources, the development and refinement of schematic figures, and the preparation of supporting materials.</p>
      </sec>
      <sec>
        <title>Authors’ contributions</title>
        <p>The author contributed solely to the article.</p>
      </sec>
      <sec>
        <title>Availability of data and materials</title>
        <p>Not applicable.</p>
      </sec>
      <sec>
        <title>AI and AI-assisted tools statement</title>
        <p>During the preparation of this manuscript, the AI tool ChatGPT (OpenAI, version GPT-5.5, released 2026-04-23) was used to assist in preparing some decorative icons and schematic elements in <xref ref-type="fig" rid="fig1">Figures 1</xref>-<xref ref-type="fig" rid="fig10">10</xref>, and GPT-6 Sol (released 2026-09-22) was used to assist in preparing decorative illustrations representing scientific and technological themes in the Graphical Abstract. These tools did not influence the study design, data collection, analysis, interpretation, or the scientific content of the work. The author takes full responsibility for the accuracy, integrity, and final content of the manuscript.</p>
      </sec>
      <sec>
        <title>Financial support and sponsorship</title>
        <p>This work was supported by Suzhou MatSource Technology Co., Ltd. (Suzhou, China) and the Gusu Laboratory of Materials (Suzhou, China), under grant number Y2501.</p>
      </sec>
      <sec>
        <title>Conflicts of interest</title>
        <p>Li, L. is an Editorial Board Member of <italic>AI Agent</italic> and is affiliated with Suzhou MatSource Technology Co., Ltd., which provided financial support for this work. Li, L. was not involved in any steps of the editorial process, including reviewers’ selection, manuscript handling, or decision-making.</p>
      </sec>
      <sec>
        <title>Ethical approval and consent to participate</title>
        <p>Not applicable.</p>
      </sec>
      <sec>
        <title>Consent for publication</title>
        <p>Not applicable.</p>
      </sec>
      <sec>
        <title>Copyright</title>
        <p>© The Author(s) 2026.</p>
      </sec>
    </sec>
    <ref-list>
      <ref id="B1">
        <label>1</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Xi</surname>
              <given-names>Z.</given-names>
            </name>
            <name>
              <surname>Chen</surname>
              <given-names>W.</given-names>
            </name>
            <name>
              <surname>Guo</surname>
              <given-names>X.</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>The rise and potential of large language model based agents: a survey</article-title>
          <source>Sci. China Inf. Sci.</source>
          <year>2025</year>
          <volume>68</volume>
          <fpage>121101</fpage>
          <pub-id pub-id-type="doi">10.1007/s11432-024-4222-0</pub-id>
        </element-citation>
      </ref>
      <ref id="B2">
        <label>2</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Yao</surname>
              <given-names>S.</given-names>
            </name>
            <name>
              <surname>Zhao</surname>
              <given-names>J.</given-names>
            </name>
            <name>
              <surname>Yu</surname>
              <given-names>D.</given-names>
            </name>
            <etal/>
          </person-group>
		  <comment>ReAct: synergizing reasoning and acting in language models. <italic>arXiv</italic> <bold>2023</bold>, arXiv:2210.03629. Available online: <uri xlink:href="https://doi.org/10.48550/arXiv.2210.03629">https://doi.org/10.48550/arXiv.2210.03629</uri> (accessed 2 November 2026)</comment>
        </element-citation>
      </ref>
      <ref id="B3">
        <label>3</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Schick</surname>
              <given-names>T.</given-names>
            </name>
            <name>
              <surname>Dwivedi-Yu</surname>
              <given-names>J.</given-names>
            </name>
            <name>
              <surname>Dessi</surname>
              <given-names>R.</given-names>
            </name>
            <etal/>
          </person-group>
          <comment>Toolformer: language models can teach themselves to use tools. In NIPS '23: Proceedings of the 37th International Conference on Neural Information Processing Systems, New Orleans, LA, USA; Curran Associates Inc.: Red Hook, NY, USA, 2023; pp 68539-51. <uri xlink:href="https://dl.acm.org/doi/10.5555/3666122.3669119">https://dl.acm.org/doi/10.5555/3666122.3669119</uri>. (accessed 2026-11-02)</comment>
        </element-citation>
      </ref>
      <ref id="B4">
        <label>4</label>
        <element-citation publication-type="conference">
          <person-group person-group-type="author">
            <name>
              <surname>Park</surname>
              <given-names>J. S.</given-names>
            </name>
            <name>
              <surname>O’Brien</surname>
              <given-names>J.</given-names>
            </name>
            <name>
              <surname>Cai</surname>
              <given-names>C. J.</given-names>
            </name>
            <name>
              <surname>Morris</surname>
              <given-names>M. R.</given-names>
            </name>
            <name>
              <surname>Liang</surname>
              <given-names>P.</given-names>
            </name>
            <name>
              <surname>Bernstein</surname>
              <given-names>M. S.</given-names>
            </name>
          </person-group>
          <comment>Generative agents: interactive simulacra of human behavior. In <italic>UIST '23: The 36th Annual ACM Symposium on User Interface Software and Technology</italic>, San Francisco CA USA; Association for Computing Machinery: New York, NY, USA, 2023; pp 1-22</comment>
          <pub-id pub-id-type="doi">10.1145/3586183.3606763</pub-id>
        </element-citation>
      </ref>
      <ref id="B5">
        <label>5</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Wang</surname>
              <given-names>L.</given-names>
            </name>
            <name>
              <surname>Ma</surname>
              <given-names>C.</given-names>
            </name>
            <name>
              <surname>Feng</surname>
              <given-names>X.</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>A survey on large language model based autonomous agents</article-title>
          <source>Front. Comput. Sci.</source>
          <year>2024</year>
          <volume>18</volume>
          <fpage>186345</fpage>
          <pub-id pub-id-type="doi">10.1007/s11704-024-40231-1</pub-id>
        </element-citation>
      </ref>
      <ref id="B6">
        <label>6</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Webb</surname>
              <given-names>T.</given-names>
            </name>
            <name>
              <surname>Mondal</surname>
              <given-names>S. S.</given-names>
            </name>
            <name>
              <surname>Momennejad</surname>
              <given-names>I.</given-names>
            </name>
          </person-group>
          <article-title>A brain-inspired agentic architecture to improve planning with LLMs</article-title>
          <source>Nat. Commun.</source>
          <year>2025</year>
          <volume>16</volume>
          <fpage>8633</fpage>
          <pub-id pub-id-type="doi">10.1038/s41467-025-63804-5</pub-id>
          <pub-id pub-id-type="pmid">41028742</pub-id>
          <pub-id pub-id-type="pmcid">PMC12485071</pub-id>
        </element-citation>
      </ref>
      <ref id="B7">
        <label>7</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Huang</surname>
              <given-names>X.</given-names>
            </name>
            <name>
              <surname>Liu</surname>
              <given-names>W.</given-names>
            </name>
            <name>
              <surname>Chen</surname>
              <given-names>X.</given-names>
            </name>
            <etal/>
          </person-group>
		  <comment>Understanding the planning of LLM agents: a survey. <italic>arXiv</italic> <bold>2024</bold>, arXiv:2402.02716. Available online: <uri xlink:href="https://doi.org/10.48550/arXiv.2402.02716">https://doi.org/10.48550/arXiv.2402.02716</uri> (accessed 2 November 2026)</comment>
        </element-citation>
      </ref>
      <ref id="B8">
        <label>8</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Wang</surname>
              <given-names>G.</given-names>
            </name>
            <name>
              <surname>Xie</surname>
              <given-names>Y.</given-names>
            </name>
            <name>
              <surname>Jiang</surname>
              <given-names>Y.</given-names>
            </name>
            <etal/>
          </person-group>
		  <comment>Voyager: an open-ended embodied agent with large language models. <italic>arXiv</italic> <bold>2023</bold>, arXiv:2305.16291. Available online: <uri xlink:href="https://doi.org/10.48550/arXiv.2305.16291">https://doi.org/10.48550/arXiv.2305.16291</uri> (accessed 2 November 2026)</comment>
        </element-citation>
      </ref>
      <ref id="B9">
        <label>9</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Trinh</surname>
              <given-names>T. H.</given-names>
            </name>
            <name>
              <surname>Wu</surname>
              <given-names>Y.</given-names>
            </name>
            <name>
              <surname>Le</surname>
              <given-names>Q. V.</given-names>
            </name>
            <name>
              <surname>He</surname>
              <given-names>H.</given-names>
            </name>
            <name>
              <surname>Luong</surname>
              <given-names>T.</given-names>
            </name>
          </person-group>
          <article-title>Solving olympiad geometry without human demonstrations</article-title>
          <source>Nature</source>
          <year>2024</year>
          <volume>625</volume>
          <fpage>476</fpage>
          <lpage>82</lpage>
          <pub-id pub-id-type="doi">10.1038/s41586-023-06747-5</pub-id>
          <pub-id pub-id-type="pmid">38233616</pub-id>
          <pub-id pub-id-type="pmcid">PMC10794143</pub-id>
        </element-citation>
      </ref>
      <ref id="B10">
        <label>10</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Polu</surname>
              <given-names>S.</given-names>
            </name>
            <name>
              <surname>Sutskever</surname>
              <given-names>I.</given-names>
            </name>
          </person-group>
		  <comment>Generative language modeling for automated theorem proving. <italic>arXiv</italic> <bold>2020</bold>, arXiv:2009.03393. Available online: <uri xlink:href="https://doi.org/10.48550/arXiv.2009.03393">https://doi.org/10.48550/arXiv.2009.03393</uri> (accessed 2 November 2026)</comment>
        </element-citation>
      </ref>
      <ref id="B11">
        <label>11</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Zheng</surname>
              <given-names>K.</given-names>
            </name>
            <name>
              <surname>Han</surname>
              <given-names>J. M.</given-names>
            </name>
            <name>
              <surname>Polu</surname>
              <given-names>S.</given-names>
            </name>
          </person-group>
          <comment>MiniF2F: a cross-system benchmark for formal olympiad-level mathematics. In <italic>International Conference on Learning Representations (ICLR)</italic>, 2022. <uri xlink:href="https://openreview.net/forum?id=9ZPegFuFTFv">https://openreview.net/forum?id=9ZPegFuFTFv</uri>. (accessed 2026-11-02)</comment>
        </element-citation>
      </ref>
      <ref id="B12">
        <label>12</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Bran</surname>
              <given-names>A. M.</given-names>
            </name>
            <name>
              <surname>Cox</surname>
              <given-names>S.</given-names>
            </name>
            <name>
              <surname>Schilter</surname>
              <given-names>O.</given-names>
            </name>
            <name>
              <surname>Baldassari</surname>
              <given-names>C.</given-names>
            </name>
            <name>
              <surname>White</surname>
              <given-names>A. D.</given-names>
            </name>
            <name>
              <surname>Schwaller</surname>
              <given-names>P.</given-names>
            </name>
          </person-group>
          <article-title>Augmenting large language models with chemistry tools</article-title>
          <source>Nat. Mach. Intell.</source>
          <year>2024</year>
          <volume>6</volume>
          <fpage>525</fpage>
          <lpage>35</lpage>
          <pub-id pub-id-type="doi">10.1038/s42256-024-00832-8</pub-id>
          <pub-id pub-id-type="pmid">38799228</pub-id>
          <pub-id pub-id-type="pmcid">PMC11116106</pub-id>
        </element-citation>
      </ref>
      <ref id="B13">
        <label>13</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Boiko</surname>
              <given-names>D. A.</given-names>
            </name>
            <name>
              <surname>MacKnight</surname>
              <given-names>R.</given-names>
            </name>
            <name>
              <surname>Kline</surname>
              <given-names>B.</given-names>
            </name>
            <name>
              <surname>Gomes</surname>
              <given-names>G.</given-names>
            </name>
          </person-group>
          <article-title>Autonomous chemical research with large language models</article-title>
          <source>Nature</source>
          <year>2023</year>
          <volume>624</volume>
          <fpage>570</fpage>
          <lpage>8</lpage>
          <pub-id pub-id-type="doi">10.1038/s41586-023-06792-0</pub-id>
          <pub-id pub-id-type="pmid">38123806</pub-id>
          <pub-id pub-id-type="pmcid">PMC10733136</pub-id>
        </element-citation>
      </ref>
      <ref id="B14">
        <label>14</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Burger</surname>
              <given-names>B.</given-names>
            </name>
            <name>
              <surname>Maffettone</surname>
              <given-names>P. M.</given-names>
            </name>
            <name>
              <surname>Gusev</surname>
              <given-names>V. V.</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>A mobile robotic chemist</article-title>
          <source>Nature</source>
          <year>2020</year>
          <volume>583</volume>
          <fpage>237</fpage>
          <lpage>41</lpage>
          <pub-id pub-id-type="doi">10.1038/s41586-020-2442-2</pub-id>
          <pub-id pub-id-type="pmid">32641813</pub-id>
        </element-citation>
      </ref>
      <ref id="B15">
        <label>15</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Zhang</surname>
              <given-names>D.</given-names>
            </name>
            <name>
              <surname>Li</surname>
              <given-names>H.</given-names>
            </name>
          </person-group>
          <comment>Digital Catalysis Platform (DigCat): a gateway to big data and AI-powered innovations in catalysis. <italic>ChemRxiv</italic> <bold>2026</bold>, Epub ahead of print</comment>
          <pub-id pub-id-type="doi">10.26434/chemrxiv-2024-9lpb9/v3</pub-id>
        </element-citation>
      </ref>
      <ref id="B16">
        <label>16</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Zhang</surname>
              <given-names>D.</given-names>
            </name>
            <name>
              <surname>Jia</surname>
              <given-names>X.</given-names>
            </name>
            <name>
              <surname>Liu</surname>
              <given-names>H.</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>Cloud synthesis: a global closed-loop feedback powered by autonomous AI-driven catalyst design agent</article-title>
          <source>AI Agent</source>
          <year>2025</year>
          <volume>1</volume>
          <fpage>2</fpage>
          <pub-id pub-id-type="doi">10.20517/aiagent.2025.02</pub-id>
        </element-citation>
      </ref>
      <ref id="B17">
        <label>17</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Zhang</surname>
              <given-names>D.</given-names>
            </name>
            <name>
              <surname>Jia</surname>
              <given-names>X.</given-names>
            </name>
            <name>
              <surname>Hung</surname>
              <given-names>T. B.</given-names>
            </name>
            <etal/>
          </person-group>
		  <comment>“DIVE” into hydrogen storage materials discovery with AI agents. <italic>arXiv</italic> <bold>2025</bold>, arXiv:2508.13251. Available online: <uri xlink:href="https://doi.org/10.48550/arXiv.2508.13251">https://doi.org/10.48550/arXiv.2508.13251</uri> (accessed 2 November 2026)</comment>
        </element-citation>
      </ref>
      <ref id="B18">
        <label>18</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Wang</surname>
              <given-names>H.</given-names>
            </name>
            <name>
              <surname>Li</surname>
              <given-names>K.</given-names>
            </name>
            <name>
              <surname>Ramsay</surname>
              <given-names>S.</given-names>
            </name>
            <name>
              <surname>Fehlis</surname>
              <given-names>Y.</given-names>
            </name>
            <name>
              <surname>Kim</surname>
              <given-names>E.</given-names>
            </name>
            <name>
              <surname>Hattrick-Simpers</surname>
              <given-names>J.</given-names>
            </name>
          </person-group>
          <article-title>Evaluating the performance and robustness of LLMs in materials science Q&amp;A and property predictions</article-title>
          <source>Digit. Discov.</source>
          <year>2025</year>
          <volume>4</volume>
          <fpage>1612</fpage>
          <lpage>24</lpage>
          <pub-id pub-id-type="doi">10.1039/d5dd00090d</pub-id>
        </element-citation>
      </ref>
      <ref id="B19">
        <label>19</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Merchant</surname>
              <given-names>A.</given-names>
            </name>
            <name>
              <surname>Batzner</surname>
              <given-names>S.</given-names>
            </name>
            <name>
              <surname>Schoenholz</surname>
              <given-names>S. S.</given-names>
            </name>
            <name>
              <surname>Aykol</surname>
              <given-names>M.</given-names>
            </name>
            <name>
              <surname>Cheon</surname>
              <given-names>G.</given-names>
            </name>
            <name>
              <surname>Cubuk</surname>
              <given-names>E. D.</given-names>
            </name>
          </person-group>
          <article-title>Scaling deep learning for materials discovery</article-title>
          <source>Nature</source>
          <year>2023</year>
          <volume>624</volume>
          <fpage>80</fpage>
          <lpage>5</lpage>
          <pub-id pub-id-type="doi">10.1038/s41586-023-06735-9</pub-id>
          <pub-id pub-id-type="pmid">38030720</pub-id>
          <pub-id pub-id-type="pmcid">PMC10700131</pub-id>
        </element-citation>
      </ref>
      <ref id="B20">
        <label>20</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Jain</surname>
              <given-names>A.</given-names>
            </name>
            <name>
              <surname>Ong</surname>
              <given-names>S. P.</given-names>
            </name>
            <name>
              <surname>Hautier</surname>
              <given-names>G.</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>Commentary: the Materials Project: a materials genome approach to accelerating materials innovation</article-title>
          <source>APL Mater.</source>
          <year>2013</year>
          <volume>1</volume>
          <fpage>011002</fpage>
          <pub-id pub-id-type="doi">10.1063/1.4812323</pub-id>
        </element-citation>
      </ref>
      <ref id="B21">
        <label>21</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Abramson</surname>
              <given-names>J.</given-names>
            </name>
            <name>
              <surname>Adler</surname>
              <given-names>J.</given-names>
            </name>
            <name>
              <surname>Dunger</surname>
              <given-names>J.</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>Accurate structure prediction of biomolecular interactions with AlphaFold 3</article-title>
          <source>Nature</source>
          <year>2024</year>
          <volume>630</volume>
          <fpage>493</fpage>
          <lpage>500</lpage>
          <pub-id pub-id-type="doi">10.1038/s41586-024-07487-w</pub-id>
          <pub-id pub-id-type="pmid">38718835</pub-id>
          <pub-id pub-id-type="pmcid">PMC11168924</pub-id>
        </element-citation>
      </ref>
      <ref id="B22">
        <label>22</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Lin</surname>
              <given-names>Z.</given-names>
            </name>
            <name>
              <surname>Akin</surname>
              <given-names>H.</given-names>
            </name>
            <name>
              <surname>Rao</surname>
              <given-names>R.</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>Evolutionary-scale prediction of atomic-level protein structure with a language model</article-title>
          <source>Science</source>
          <year>2023</year>
          <volume>379</volume>
          <fpage>1123</fpage>
          <lpage>30</lpage>
          <pub-id pub-id-type="doi">10.1126/science.ade2574</pub-id>
          <pub-id pub-id-type="pmid">36927031</pub-id>
        </element-citation>
      </ref>
      <ref id="B23">
        <label>23</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Jumper</surname>
              <given-names>J.</given-names>
            </name>
            <name>
              <surname>Evans</surname>
              <given-names>R.</given-names>
            </name>
            <name>
              <surname>Pritzel</surname>
              <given-names>A.</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>Highly accurate protein structure prediction with AlphaFold</article-title>
          <source>Nature</source>
          <year>2021</year>
          <volume>596</volume>
          <fpage>583</fpage>
          <lpage>9</lpage>
          <pub-id pub-id-type="doi">10.1038/s41586-021-03819-2</pub-id>
          <pub-id pub-id-type="pmid">34265844</pub-id>
          <pub-id pub-id-type="pmcid">PMC8371605</pub-id>
        </element-citation>
      </ref>
      <ref id="B24">
        <label>24</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Huang</surname>
              <given-names>K.</given-names>
            </name>
            <name>
              <surname>Zhang</surname>
              <given-names>S.</given-names>
            </name>
            <name>
              <surname>Wang</surname>
              <given-names>H.</given-names>
            </name>
            <etal/>
          </person-group>
          <comment>Biomni: a general-purpose biomedical AI agent. <italic>bioRxiv</italic> <bold>2025</bold>, Epub ahead of print</comment>
          <pub-id pub-id-type="doi">10.1101/2025.05.30.656746</pub-id>
          <pub-id pub-id-type="pmid">40501924</pub-id>
          <pub-id pub-id-type="pmcid">PMC12157518</pub-id>
        </element-citation>
      </ref>
      <ref id="B25">
        <label>25</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Roohani</surname>
              <given-names>Y.</given-names>
            </name>
            <name>
              <surname>Lee</surname>
              <given-names>A.</given-names>
            </name>
            <name>
              <surname>Huang</surname>
              <given-names>Q.</given-names>
            </name>
            <etal/>
          </person-group>
          <comment>BioDiscoveryAgent: an AI agent for designing genetic perturbation experiments. In <italic>International Conference on Learning Representations (ICLR)</italic>, 2025. <uri xlink:href="https://proceedings.iclr.cc/paper_files/paper/2025/hash/4252dc94531833029000f85dc5fac792-Abstract-Conference.html">https://proceedings.iclr.cc/paper_files/paper/2025/hash/4252dc94531833029000f85dc5fac792-Abstract-Conference.html</uri>. (accessed 2026-11-02)</comment>
        </element-citation>
      </ref>
      <ref id="B26">
        <label>26</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Gridach</surname>
              <given-names>M.</given-names>
            </name>
            <name>
              <surname>Nanavati</surname>
              <given-names>J.</given-names>
            </name>
			<name>
              <surname>Mack</surname>
              <given-names>C.</given-names>
            </name>
            <name>
              <surname>El Abidine</surname>
              <given-names>K. Z.</given-names>
            </name>
            <name>
              <surname>Mendes</surname>
              <given-names>L.</given-names>
            </name>
          </person-group>
          <comment>Agentic AI for scientific discovery: a survey of progress, challenges, and future directions. In <italic>ICLR 2025 Workshops: AgenticAI</italic>, 2025. <uri xlink:href="https://openreview.net/forum?id=TyCYakX9BD">https://openreview.net/forum?id=TyCYakX9BD</uri>. (accessed 2026-11-02)</comment>
        </element-citation>
      </ref>
      <ref id="B27">
        <label>27</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Vriza</surname>
              <given-names>A.</given-names>
            </name>
            <name>
              <surname>Prince</surname>
              <given-names>M. H.</given-names>
            </name>
            <name>
              <surname>Zhou</surname>
              <given-names>T.</given-names>
            </name>
            <name>
              <surname>Chan</surname>
              <given-names>H.</given-names>
            </name>
            <name>
              <surname>Cherukara</surname>
              <given-names>M. J.</given-names>
            </name>
          </person-group>
          <article-title>Operating advanced scientific instruments with AI agents that learn on the job</article-title>
          <source>npj Comput. Mater.</source>
          <year>2026</year>
          <volume>12</volume>
          <fpage>160</fpage>
          <pub-id pub-id-type="doi">10.1038/s41524-026-02005-0</pub-id>
        </element-citation>
      </ref>
      <ref id="B28">
        <label>28</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Rolnick</surname>
              <given-names>D.</given-names>
            </name>
            <name>
              <surname>Donti</surname>
              <given-names>P. L.</given-names>
            </name>
            <name>
              <surname>Kaack</surname>
              <given-names>L. H.</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>Tackling climate change with machine learning</article-title>
          <source>ACM Comput. Surv.</source>
          <year>2022</year>
          <volume>55</volume>
          <fpage>1</fpage>
          <lpage>96</lpage>
          <pub-id pub-id-type="doi">10.1145/3485128</pub-id>
        </element-citation>
      </ref>
      <ref id="B29">
        <label>29</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Vázquez-canteli</surname>
              <given-names>J. R.</given-names>
            </name>
            <name>
              <surname>Nagy</surname>
              <given-names>Z.</given-names>
            </name>
          </person-group>
          <article-title>Reinforcement learning for demand response: a review of algorithms and modeling techniques</article-title>
          <source>Appl. Energy</source>
          <year>2019</year>
          <volume>235</volume>
          <fpage>1072</fpage>
          <lpage>89</lpage>
          <pub-id pub-id-type="doi">10.1016/j.apenergy.2018.11.002</pub-id>
        </element-citation>
      </ref>
      <ref id="B30">
        <label>30</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Kao</surname>
              <given-names>C. H.</given-names>
            </name>
            <name>
              <surname>Zhao</surname>
              <given-names>W.</given-names>
            </name>
            <name>
              <surname>Revankar</surname>
              <given-names>S.</given-names>
            </name>
            <etal/>
          </person-group>
          <comment>Towards LLM agents for Earth observation. In <italic>ICML 2025 Workshop on TerraBytes: Towards Global Datasets and Models for Earth Observation</italic>, 2025. <uri xlink:href="https://arxiv.org/abs/2504.12110v2">https://arxiv.org/abs/2504.12110v2</uri>. (accessed 2026-11-02)</comment>
        </element-citation>
      </ref>
      <ref id="B31">
        <label>31</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Feng</surname>
              <given-names>P.</given-names>
            </name>
            <name>
              <surname>Lv</surname>
              <given-names>Z.</given-names>
            </name>
            <name>
              <surname>Ye</surname>
              <given-names>J.</given-names>
            </name>
            <etal/>
          </person-group>
		  <comment>Earth-Agent: unlocking the full landscape of Earth observation with agents. <italic>arXiv</italic> <bold>2026</bold>, arXiv:2509.23141. Available online: <uri xlink:href="https://doi.org/10.48550/arXiv.2509.23141">https://doi.org/10.48550/arXiv.2509.23141</uri> (accessed 2 November 2026)</comment>
        </element-citation>
      </ref>
      <ref id="B32">
        <label>32</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Zitkovich</surname>
              <given-names>B.</given-names>
            </name>
            <name>
              <surname>Yu</surname>
              <given-names>T.</given-names>
            </name>
            <name>
              <surname>Xu</surname>
              <given-names>S.</given-names>
            </name>
            <etal/>
          </person-group>
          <comment>RT-2: vision-language-action models transfer web knowledge to robotic control. In <italic>Proceedings of the 7th Conference on Robot Learning</italic>, 2023; Vol. 229, pp 2165-83. <uri xlink:href="https://proceedings.mlr.press/v229/zitkovich23a.html">https://proceedings.mlr.press/v229/zitkovich23a.html</uri>. (accessed 2026-11-02)</comment>
        </element-citation>
      </ref>
      <ref id="B33">
        <label>33</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Kim</surname>
              <given-names>M. J.</given-names>
            </name>
            <name>
              <surname>Pertsch</surname>
              <given-names>K.</given-names>
            </name>
            <name>
              <surname>Karamcheti</surname>
              <given-names>S.</given-names>
            </name>
            <etal/>
          </person-group>
		  <comment>OpenVLA: an open-source vision-language-action model. <italic>arXiv</italic> <bold>2024</bold>, arXiv:2406.09246. Available online: <uri xlink:href="https://doi.org/10.48550/arXiv.2406.09246">https://doi.org/10.48550/arXiv.2406.09246</uri> (accessed 2 November 2026)</comment>
        </element-citation>
      </ref>
      <ref id="B34">
        <label>34</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Black</surname>
              <given-names>K.</given-names>
            </name>
            <name>
              <surname>Brown</surname>
              <given-names>N.</given-names>
            </name>
            <name>
              <surname>Driess</surname>
              <given-names>D.</given-names>
            </name>
            <etal/>
          </person-group>
		  <comment>π0: a vision-language-action flow model for general robot control. <italic>arXiv</italic> <bold>2026</bold>, arXiv:2410.24164. Available online: <uri xlink:href="https://doi.org/10.48550/arXiv.2410.24164">https://doi.org/10.48550/arXiv.2410.24164</uri> (accessed 2 November 2026)</comment>
        </element-citation>
      </ref>
      <ref id="B35">
        <label>35</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Greshake</surname>
              <given-names>K.</given-names>
            </name>
            <name>
              <surname>Abdelnabi</surname>
              <given-names>S.</given-names>
            </name>
            <name>
              <surname>Mishra</surname>
              <given-names>S.</given-names>
            </name>
            <name>
              <surname>Endres</surname>
              <given-names>C.</given-names>
            </name>
            <name>
              <surname>Holz</surname>
              <given-names>T.</given-names>
            </name>
            <name>
              <surname>Fritz</surname>
              <given-names>M.</given-names>
            </name>
          </person-group>
		  <comment>Not what you’ve signed up for: compromising real-world LLM-integrated applications with indirect prompt injection. <italic>arXiv</italic> <bold>2023</bold>, arXiv:2302.12173. Available online: <uri xlink:href="https://doi.org/10.48550/arXiv.2302.12173">https://doi.org/10.48550/arXiv.2302.12173</uri> (accessed 2 November 2026)</comment>
        </element-citation>
      </ref>
      <ref id="B36">
        <label>36</label>
        <element-citation publication-type="conference">
          <person-group person-group-type="author">
            <name>
              <surname>Perez</surname>
              <given-names>E.</given-names>
            </name>
            <name>
              <surname>Huang</surname>
              <given-names>S.</given-names>
            </name>
            <name>
              <surname>Song</surname>
              <given-names>F.</given-names>
            </name>
            <etal/>
          </person-group>
          <comment>Red teaming language models with language models. In <italic>Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing</italic>, Abu Dhabi, United Arab Emirates, December, 2022; Association for Computational Linguistics: Stroudsburg, PA, USA, 2022; pp 3419-48</comment>
          <pub-id pub-id-type="doi">10.18653/v1/2022.emnlp-main.225</pub-id>
        </element-citation>
      </ref>
      <ref id="B37">
        <label>37</label>
        <element-citation publication-type="web">
          <person-group person-group-type="author">
            <name>
              <surname>Zhang</surname>
              <given-names>H.</given-names>
            </name>
            <name>
              <surname>Huang</surname>
              <given-names>J.</given-names>
            </name>
            <name>
              <surname>Mei</surname>
              <given-names>K.</given-names>
            </name>
            <etal/>
          </person-group>
          <comment>Agent Security Bench (ASB): formalizing and benchmarking attacks and defenses in LLM-based agents. In <italic>International Conference on Learning Representations (ICLR)</italic>, 2025. <uri xlink:href="https://openreview.net/forum?id=V4y0CpX4hK">https://openreview.net/forum?id=V4y0CpX4hK</uri>. (accessed 2026-11-02)</comment>
        </element-citation>
      </ref>
    </ref-list>
  </back>
</article>
