Agent Discovery in Internet of Agents: Challenges and Solutions

TL;DR

This paper tackles the problem of discovering capabilities in the Internet of Agents (IoA), where heterogeneous agents must efficiently identify suitable collaborators under dynamic tasks. It proposes a two-stage framework: autonomous capability announcement and task-driven capability discovery, augmented by a semantic-driven scheme that includes language-model-based profiling, scalable indexing via discrete codes, and memory-enhanced continual discovery. The approach demonstrates improved discovery performance and scalability in simulations, outperforming traditional baselines and maintaining effectiveness as agent populations grow to thousands. The work highlights a pragmatic roadmap with cross-domain collaboration, security/privacy, and economic models as key avenues to enable robust, scalable IoA ecosystems.

Abstract

Rapid advances in large language models and agentic AI are driving the emergence of the Internet of Agents (IoA), a paradigm where billions of autonomous software and embodied agents interact, coordinate, and collaborate to accomplish complex tasks. A key prerequisite for such large-scale collaboration is agent capability discovery, where agents identify, advertise, and match one another's capabilities under dynamic tasks. Agent's capability in IoA is inherently heterogeneous and context-dependent, raising challenges in capability representation, scalable discovery, and long-term performance. To address these issues, this paper introduces a novel two-stage capability discovery framework. The first stage, autonomous capability announcement, allows agents to credibly publish machine-interpretable descriptions of their abilities. The second stage, task-driven capability discovery, enables context-aware search, ranking, and composition to locate and assemble suitable agents for specific tasks. Building on this framework, we propose a novel scheme that integrates semantic capability modeling, scalable and updatable indexing, and memory-enhanced continual discovery. Simulation results demonstrate that our approach enhances discovery performance and scalability. Finally, we outline a research roadmap and highlight open problems and promising directions for future IoA.

Figures (5)

• Figure 1: IoA Architecture for Capability-Driven Autonomous Collaboration.
• Figure 2: Overview of agent discovery in IoA. Each agent advertises its capabilities (i.e., what can I offer) and discovers others based on task demands (i.e., what do I need), enabling autonomous collaboration.
• Figure 3: Illustration of semantic-driven capability discovery solutions in IoA, including: (a) language model-powered semantic profiling of agents, (b) scalable and efficient indexing, and (c) memory-enhanced continual discovery.
• Figure 4: Comparison of agent discovery performance between ours and benchmarks.
• Figure 5: Comparison of discovery performance in three schemes under varying numbers of agents.