cs.AIMar 15, 2026

Autonomous Agents Coordinating Distributed Discovery Through Emergent Artifact Exchange

Authors: Fiona Y. Wang, Lee Marom, Subhadeep Pal, Rachel K. Luu, Wei Lu, Jaime A. Berkovich, Markus J. Buehler

Organizations: Laboratory for Atomistic and Molecular Mechanics (LAMM), Massachusetts Institute of Technology, Cambridge, MA 02139, USA · Department of Biological Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA · Department of Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA · Department of Civil and Environmental Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA · 

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

We present ScienceClaw + Infinite, a framework for autonomous scientific investigation in which independent agents conduct research without central coordination, and any contributor can deploy new agents into a shared ecosystem. The system is built around three components: an extensible registry of over 300 interoperable scientific skills, an artifact layer that preserves full computational lineage as a directed acyclic graph (DAG), and a structured platform for agent-based scientific discourse with provenance-aware governance. Agents select and chain tools based on their scientific profiles, produce immutable artifacts with typed metadata and parent lineage, and broadcast unsatisfied information needs to a shared global index. The ArtifactReactor enables plannerless coordination: peer agents discover and fulfill open needs through pressure-based scoring, while schema-overlap matching triggers multi-parent synthesis across independent analyses. An autonomous mutation layer actively prunes the expanding artifact DAG to resolve conflicting or redundant workflows, while persistent memory allows agents to continuously build upon complex epistemic states across multiple cycles. Infinite converts these outputs into auditable scientific records through structured posts, provenance views, and machine-readable discourse relations, with community feedback steering subsequent investigation cycles. Across four autonomous investigations, peptide design for the somatostatin receptor SSTR2, lightweight impact-resistant ceramic screening, cross-domain resonance bridging biology, materials, and music, and formal analogy construction between urban morphology and grain-boundary evolution, the framework demonstrates heterogeneous tool chaining, emergent convergence among independently operating agents, and traceable reasoning from raw computation to published finding.

Figures & tables

Figure 1: ScienceClaw + Infinite ecosystem loop. (1) ScienceClaw : Agents invoke domain skills. (2) Computations: Skills execute, producing raw results. (3) Artifacts: Results wrapped as immutable, addressable records with parent lineage, accumulated in a shared DAG; need signals broadcast to a global index for peer discovery. (4) Figures: A plot agent renders visualizations from the artifact graph. (5) Infinite Interaction: Findings published as structured posts with evidence surfaces and artifact provenance. (6) Feedback: Votes, actions, and redirects from agent and human peers feed back to the ArtifactReactor’s pressure scorer, which biases the next cycle toward high-impact, under-explored directions.
Case StudyAgentsToolsArtifactsSynthesisAvg DAG
Protein Design1023177572.15
Materials Discovery81073222.25
Resonance Landscape1312159192.00
Formal Analogy92352252.00
Table 1: Comprehensive quantitative evaluation of autonomous case studies. Agents: autonomous agents participating in the investigation. Tools: unique computational tools invoked across all agents. Artifacts: total number of unique computational artifacts generated (deduplicated by artifact ID). Synthesis: number of artifacts that combine, integrate, or transform data and results across multiple sources to produce new aggregated outputs. Includes: ranked candidate lists aggregating multiple screening results, knowledge graphs and network visualizations combining concepts from multiple papers, statistical models fitted to integrated datasets, integration figures merging tool outputs, and explicit synthesis operations that produce summary artifacts. Avg DAG: average dependency chain depth calculated from artifact parent relationships across all artifacts with parents. All scientific investigation processes were conducted without human redirects or interventions. Report generation and figure creation were executed by human for materials and resonance case studies.
Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Figure S1: Agent Interaction Network: Protein Design. Nodes represent autonomous agents (large colored circles, labeled with comment count) and their individual comments (small satellite nodes). Edges encode three interaction patterns: mention/reply edges (gray solid lines) for direct agent discussion; artifact-passed edges (orange dashed arrows) labeled with artifact types (pubmed_results, sequence_alignment, admet_prediction, protein_data, synthesis) showing cross-agent dependencies where one agent’s tool output becomes input to another’s analysis; and comment satellites (light gray dashed lines). Densely clustered central core of 10 agents reveals continuous multi-stage artifact integration across 23 computational tools, producing 177 artifacts with 57 synthesis operations (32% synthesis density).
Figure S2: Agent Interaction Network: Materials Science. Nodes represent autonomous agents (large colored circles, labeled with comment count) and their individual comments (small satellite nodes). Edges encode three interaction patterns: mention/reply edges (gray solid lines) for direct agent discussion; artifact-passed edges (orange dashed arrows) labeled with artifact types (pubmed_results, materials_data, ml_prediction, synthesis) showing cross-agent dependencies; and comment satellites (light gray dashed lines). Compact core of 8 agents with concentrated, heavy (arrow weight scales with artifact counts) artifact flows between agent pairs (LitMiner, MaterialsSynth, CandidateRanker, Structural Analyst) indicates tightly coordinated multi-stage refinement using 10 computational tools, producing 73 artifacts with 22 synthesis operations (30% synthesis density).
Figure S3: Agent Interaction Network: Resonance Landscape. Nodes represent autonomous agents (large colored circles, labeled with comment count) and their individual comments (small satellite nodes). Edges encode three interaction patterns: mention/reply edges (gray solid lines) for direct agent discussion; artifact-passed edges (orange dashed arrows) labeled with artifact types (synthesis, computational_analysis, rdkit_properties, accumulated_results) showing cross-agent dependencies; and comment satellites (light gray dashed lines). Radial branching structure with the central agents as central hubs reflects 13 agents conducting parallel parameter sweeps across 12 computational tools that reconverge during synthesis, producing 159 artifacts with 19 synthesis operations (12% synthesis density).
Figure S4: Agent Interaction Network: Formal Analogy. Nodes represent autonomous agents (large colored circles, labeled with comment count) and their individual comments (small satellite nodes). Edges encode three interaction patterns: mention/reply edges (gray solid lines) for direct agent discussion; artifact-passed edges (orange dashed arrows and triangular synthesis markers) labeled with artifact types (materials_data, figure, ml_prediction, pubmed_results, network_data) showing cross-agent dependencies; and comment satellites (light gray dashed lines). Dense synthesis hub structure (orange triangular markers) centered on GrowthRuleAnalyst and SynthBotRun indicates frequent multi-agent knowledge consolidation: 9 agents used all 23 computational tools, producing 52 artifacts with 25 synthesis operations (48% synthesis density, highest among all studies).
More in this topic

Autonomous Scientific Discovery

Autonomous scientific discovery develops AI systems that generate, test, and refine scientific hypotheses using data, computation, and experiments.

See all 128 papers

Explore similar work

CardsList

  1. From Agent-Only Social Networks to Autonomous Scientific Research: Lessons from OpenClaw and Moltbook, and the Architecture of ClawdLab and Beach.Science

    Feb 23, 2026Lukas Weidener, Marko Brkić, Phillip Lee +3Autonomous Research AgentsAI Agents for Scientific Discovery

  2. Clarus: Coordinating Autonomous Research Agents toward Web-Scale Scientific Collaboration

    Jun 29, 2026Zihan Guo, Zeyi Chen, Zhiyu Chen +15Multi-Agent OrchestrationMulti-Agent Collaboration

    Sun Yat-sen University · Shanghai Innovation Institute · Shanghai Artificial Intelligence Laboratory +4

  3. Artifact-centered Claim-aware Observability for Autonomous Scientific Agents

    Aug 18, 2026Xiangyu Yin, Ming Du, Michael H. Prince +1AI Agent AuditingAI Agent Monitoring

    Advanced Photon Source, Argonne National Laboratory, Lemont IL, USA