Self-Evolving Agents as Dynamic Graph Transformation: A Survey and New Perspective
06:00 · August 20, 2026 · arXiv cs.AI RSS
Large language model (LLM)-based agents are increasingly becoming self-evolving systems that persist across interactions, maintain memories, use tools, acquire skills, refine workflows, and coordinate with other agents. These capabilities make agent states structural and dynamic: entities, relations, attributes, dependencies, and execution structures change with new evidence, feedback, and environmental conditions. Existing graph-agent surveys typically treat graphs as support structures for agent functions rather than as evolving substrates, while self-evolving-agent surveys focus on agent-level mechanisms and rarely discuss graph topology evolution. Thus, the coupling between evolving agent state and dynamic graph topology remains underexplored. This survey connects these two research lines by framing \textit{agent evolution as dynamic graph transformation}. We model agent state as a dynamic graph, where memories, tools, skills, workflows, and inter-agent relations are represented as typed nodes, edges, and subgraphs updated through schema-constrained rewrites. Based on this formulation, we organize existing dynamic-graph-based methods for self-evolving agents into four taxonomies: node/feature evolution, edge/topology evolution, subgraph activation, and cross-component co-evolution. Building on this taxonomy, we propose dynamic graph learning as reusable infrastructure for self-evolving agents and map nine dynamic-graph-learning subfields to agent-evolution capabilities, discussing their adaptations and possible failure modes. Finally, we discuss five types of graph-aware evaluation and governance protocols from a dynamic-graph perspective, which complement end-task evaluation. The goal is to provide a compact structural lens for designing and governing self-evolving agents.
Summary
Large language model-based agents now operate as persistent systems that accumulate memories, invoke tools, acquire skills, refine workflows, and coordinate across multiple instances. These activities produce states that are both non-stationary and structurally coupled: new evidence, feedback, or environmental change can alter entities, relations, attributes, and execution dependencies at once. Existing surveys have examined either static graphs as auxiliary structures for agent functions or agent-level adaptation mechanisms in isolation, leaving the joint evolution of agent state and graph topology largely unaddressed.
This survey closes the gap by treating agent evolution itself as dynamic graph transformation. Agent components—memories, tools, skills, workflows, and inter-agent relations—are represented as typed nodes, edges, and subgraphs whose updates follow schema-constrained rewrites. The resulting model supplies a shared formal space in which heterogeneous evolution processes can be compared and their downstream effects traced.
Existing methods are organized into four transformation patterns: node and feature evolution, edge and topology evolution, subgraph activation, and cross-component co-evolution. Building on this taxonomy, the authors map nine established subfields of dynamic graph learning onto the corresponding agent capabilities, noting the adaptations required and the failure modes each subfield may introduce when reused as agent infrastructure.
The same structural view supports five complementary evaluation and governance protocols: leakage-free temporal assessment, privacy and deletion verification, safety monitoring of propagating updates, rollback analysis, and audit through dependency graphs. Together these elements offer a compact lens for designing self-evolving agents whose internal changes remain observable, reversible, and subject to governance.
Why it matters
The paper provides foundational research on making autonomous AI agents auditable, safe, and transparent through dynamic graph modeling. This aligns strongly with the Dutch and EU focus on ethical AI and regulatory compliance, offering advanced researchers actionable frameworks for building governable agentic systems.
