Agentnetes

Zero to a Self-Organizing AI Agency. On Demand. Our hackathon is Zero to Agent? We are taking it further. It's Kubernetes for AI Agent managed by agents. Type a goal. Agentnetes assembles a self-organizing team of specialist AI agents, each running in its own Firecracker microVM sandbox, that explore your codebase, write code, run tests, fix failures, and deliver together. No hardcoded roles. No sequential bottlenecks. One goal. A full AI agency. On demand. The root agent reads your repo, invents the right team, and spawns them in parallel. A provider task gets a Scout, Engineer, Tester, and Packager. A security audit gets a completely different team. Roles are fully emergent. ▎ Built on four research foundations: the RLM Pattern (MIT CSAIL) for context-in-filesystem, the AutoResearch Loop (Karpathy) for test-driven iteration, a Two-Tool MCP surface (search + execute, ~1000 tokens), and the A2A Protocol so every agent is a discoverable service.