Cisco Outshift的高管Vijoy Pandey阐述了实现分布式人工超级智能的关键技术方向。[1]根据其观点，当前单一AI智能体已具备足够的能力，瓶颈在于缺乏协调机制——"智能已经存在，缺失的是将陌生人转变为一个团队的连接组织。"[1]为此，Outshift开发了一套系统性解决方案，通过建立语义层和连接层，使多个AI智能体能够共享意图、上下文和推理能力。[1]
多智能体系统当前的可靠性问题严重，失败率在41%至87%之间。[1]Outshift的研究表明，在采用协调协议后，非结构化团体的决策成功率可从约33%大幅提升至93%。[1]为支持这一框架的推广，Outshift开发了AGNTCY（Linux Foundation下的开源项目）和Mycelium（开源协调层）等工具，同时推出了CASA安全框架。[1]Pandey指出，约90%的情况下，智能体甚至无法确认自己是否被授权执行分配的任务。[1]对于企业的实际应用，Outshift建议从跨越3-4个团队的单一工作流开始试验多智能体协作系统。[1]
Vijoy Pandey, an executive at Cisco Outshift, has outlined a technical roadmap for advancing from isolated artificial intelligence systems toward distributed artificial superintelligence by establishing semantic and connectivity layers that enable multiple AI agents to collaborate effectively.[1] According to Pandey, "The intelligence is already there. What is missing is the connective tissue that turns four strangers into one team."[1] The framework relies on agents sharing intent, context, and reasoning capabilities across different domains and systems, a capability that current multi-agent systems struggle to achieve, with failure rates ranging between 41% and 87%.[1]
Cisco Outshift has developed several tools to address this challenge, including AGNTCY, an open-source project under the Linux Foundation, and Mycelium, an open coordination layer that organizations can deploy independently.[1] The company has also created CASA, a security framework designed to govern agent interactions.[1] Testing of the coordination protocol has demonstrated significant improvements: when applied to unstructured teams, decision-making success rates increased from approximately 33% to 93%.[1] However, a fundamental obstacle persists in current agent deployment; as Pandey noted, "Roughly 90% of the time, an agent has no way to confirm it is even cleared for the job it was handed."[1] For organizations beginning to implement this approach, Outshift recommends starting with a single workflow that spans three to four teams to pilot the technology.[1]
