AI agents are getting faster, smarter, and more autonomous. But they still work in silos. FLock has just released a new paper, laying out the vision of AgentaNet. A decentralized swarm architecture that enables agents to collaborate autonomously and securely.

2/ Most agent systems today rely on centralized control or static, closed-agent setups. This leads to 4 core problems: - No standard for agent discovery - No secure, contextual communication - Inefficient semantic exchange - No trustless onboarding AgentaNet aims to solve all of this.

3/ Feature 1: Trustworthy communication & agent profiling AgentaNet verifies what agents can actually do via benchmarking—no self-declared claims. Agents then "gossip" encrypted capability metadata using a P2P protocol. Temporary, task-scoped encrypted channels ensure privacy. When the job ends, all data is wiped.

4/ Feature 2: Decentralized task coordination AgentaNet defines modular agent roles: - Organizers parse and distribute tasks - Coordinators manage workflows - Executors perform them It supports agent-initiated tasks, decentralized negotiation, flexible workflows, and embedded smart contract guardrails to align behavior.

6/ Our paper was presented at the CFAgentic Workshop at ICML 2025, the leading global conference for machine learning and AI research.

8/ Dive into the full paper: