Autogenesis: A Self-Evolving Agent Protocol

Recent advances in LLM based agent systems have shown promise in tackling complex, long horizon tasks. However, existing agent protocols (e.g., A2A and MCP) under specify cross entity lifecycle and context management, version tracking, and evolution safe update interfaces, which encourages monolithic compositions and brittle glue code. We introduce AUTOGENESIS PRO- TOCOL (AGP), a self evolution protocol that decouples what evolves from how evolution occurs. Its Resource Substrate Protocol Layer (RSPL) models prompts, agents, tools, environments, and memory as protocol registered resources1 with explicit state, lifecycle, and versioned interfaces. Its Self Evolution Protocol Layer (SEPL) specifies a closed loop operator interface for proposing, assessing, and committing improvements with auditable lineage and rollback. Building on AGP, we present AUTOGENESIS SYSTEM (AGS), a self-evolving multi-agent system that dynamically instantiates, retrieves, and refines protocol-registered resources during execution. We evaluate AGS on multiple challenging benchmarks that require long horizon planning and tool use across heterogeneous resources.

Introduction. Recent advances in LLM-based agent systems have demonstrated significant potential in tackling complex, longhorizon tasks. However, static agent designs often prove insufficient when facing the diversity and stochasticity of real-world environments. To overcome this limitation, endowing agents with self-evolution capabilities—enabling them to automatically adjust strategies, refine instructions, and update tools based on environmental feedback—has emerged as a critical avenue for achieving robust autonomy. This transition from predefined execution to dynamic adaptation represents a fundamental shift in agentic system design. Despite the growing interest in self-evolving agents, implementations remain largely fragmented and ad hoc. Existing systems often lack shared standards, rendering the evolution process neither composable nor auditable. Developers are frequently forced to rely on brittle glue code, leading to monolithic architectures that are difficult to maintain.

Discussion / Conclusion. We presented AGP, a two-layer self-evolution protocol that decouples what evolves from how evolution occurs. The Resource Substrate Protocol Layer (RSPL) models prompts, agents, tools, environments, and memory as first-class, versioned resources with explicit lifecycle and interface con- Autogenesis: A Self-Evolving Agent Protocol