Cisco Rolls Out MyAgent Across 90,000 Employees
On August 27, 2026, Cisco said it was rolling out its MyAgent AI system to 90,000 employees, giving them access to approved large language models, agents, and enterprise data through Circuit. The company described supervised workflows across Outlook, Webex, Jira, and SharePoint, while Fortune reported that the system routes requests among models to limit frontier-model token use and relies partly on on-premises infrastructure.
Cisco is rolling out MyAgent to its 90,000-person workforce, extending a personalized enterprise AI agent across the company. In an August 27 company post, Cisco described MyAgent as a system for accessing approved large language models, agents, and enterprise data through Circuit, its secure, governed, multi-model AI platform.
The deployment follows reporting by Fortune in July that Cisco was beginning a company-wide agent rollout at the start of its fiscal year. Fortune reported that each employee would receive a personalized assistant able to handle tasks, answer questions, and route requests to an appropriate AI model.
Supervised workflows across business applications
Cisco's post describes MyAgent as operating across Outlook, Webex, Jira, SharePoint, and other applications. Employees can provide an objective, context, and desired outcome, after which the system coordinates workflow steps, according to the post. Cisco characterizes those workflows as supervised autonomous execution, with human oversight retained over the work.
The company also describes persistent memory as a core capability. According to Cisco, MyAgent can retain user preferences, prior interactions, and context over time, supporting continuity across interactions and workflows. PYMNTS, citing Cisco's post, reported that the system is designed to remain connected to approved applications and enterprise data while being securely accessible remotely.
Model routing and deployment economics
Fortune reported that Cisco's architecture dynamically chooses a model for a task rather than defaulting to frontier models. Cisco CFO Mark Patterson told Fortune, "It's not going to burn a whole bunch of tokens with frontier models." He added that the system identifies "which tool is most effective and most efficient."
Patterson also told Fortune that much of the underlying infrastructure is built on-premises, which he said gives Cisco more control over costs and data. That architecture matters because agentic workflows can use substantially more inference capacity than single-turn chat: they may plan, call tools, retrieve data, and process intermediate outputs before returning a result.
For enterprise AI teams, Cisco's deployment is a notable test of whether governed multi-model routing, application connectors, and durable user context can be operated at workforce scale. Companies pursuing comparable deployments commonly face engineering tradeoffs among authorization boundaries, auditability, retrieval quality, tool reliability, latency, and inference spend. The supplied sources establish the workflow and governance design, but do not report benchmarks for task success rates, cost per workflow, adoption, or productivity outcomes.
Cisco's executive vice president of operations wrote that the internal deployment could serve as a blueprint for customers and partners implementing enterprise AI. The available announcements do not disclose which third-party models are approved, how model-selection decisions are evaluated, or the controls used to govern actions in individual connected applications.
Key Points
- 1Cisco is extending MyAgent to 90,000 employees, making workforce-scale governed AI-agent deployment the central reported event.
- 2Circuit combines approved models, enterprise data, application connectors, and supervised workflows, concentrating governance requirements in the agent platform layer.
- 3Comparable enterprise agent deployments commonly require careful tradeoffs among model-routing cost, tool reliability, authorization controls, auditability, and output quality.
Scoring Rationale
Cisco's deployment is a notable real-world test of AI agents, multi-model routing, and governed enterprise-data access across a 90,000-person workforce. It offers relevant architectural signals for practitioners, although the supplied sources do not report outcome benchmarks, adoption metrics, or detailed technical controls.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
