蚂蚁集团开源Avernet，让人与智能体像组织一样高效协作

TL;DR - Ant Group has open-sourced Avernet (Apache 2.0), a multi-agent collaboration infrastructure whose first community release focuses on an "agent collaboration network" for discovery, consensus, cross-team coordination and governance between humans and heterogeneous agents. It matters because it targets the organizational/integration bottleneck in enterprise agent deployment rather than single-agent capability.

• Framed around four enterprise pain points — agents that are hard to find, misaligned across parties, dependent on manual handoffs, and unable to retain project experience — since privacy, compliance and business boundaries prevent centralizing everything in one "super agent."
• The community release lets agents join directly or plug in from existing platforms, then be discovered, invited, assigned tasks and return results in a unified environment, with multi-party consensus on key outputs plus observation/recording so collaboration experience is reusable. It is not tied to a single model or agent engine.
• Security governance is the stated first priority: identity authentication, access authorization, permission control, lifecycle management and some protections are partially open, answering "who is it, what can it see, what can it do." Audit trails, observability/evaluation, memory and continuous optimization, servitization and container cluster management are deferred to later versions.
• Ant reports internal deployment across 12 core business segments as of 2026-07-31 with agent task completion rates stably above 90% — vendor-provided figures, as the article notes it was supplied by Ant and republished with authorization.