跳到正文
Rohan Paul· @rohanpaul_ai · X·· 2026-08-27AI 评分41
AI 导读

思科为9万名员工部署了名为MyAgent的个性化AI智能体,背后由800多个子智能体支撑。为控制成本,50%-60%请求走开源权重模型,20%-30%走软件自动化,仅少量调用前沿模型。该智能体基于思科自2023年起构建的基础设施,通过权限化企业连接器获取员工专属上下文,外部操作需人工批准。

正文 · 原文

Cisco gave 90,000 employees personalized AI agents backed by more than 800 subagents.

For cost control, 50%-60% of requests go to open-weight models, 20%-30% to software automation, and only a small remainder goes out to a frontier model.

The agent, called MyAgent, sits on infrastructure Cisco has built since 2023, when its internal assistant initially ran on Azure OpenAI.

Now each agent pulls employee-specific context through permissioned enterprise connectors and can call more than 800 backend subagents.

Those subagents can track sales-forecast variances or send anomaly notifications, while the personal agents remain isolated from other employees' agents.

External actions still require explicit human approval, and a policy server can block destructive operations or prevent Cisco data from training third-party models.

Cisco had already built multi-model routing across Azure OpenAI, Claude, Gemini and its own Deep Network Model, so MyAgent extends an existing orchestration layer.

来源:Rohan Paul · x.com