Cisco MyAgent: 90,000 Employees Get a Personal AI Agent
On August 27, 2026, Cisco did something most enterprises are still debating in boardrooms: it deployed a personal AI agent to every single one of its 90,000 employees, not as a pilot, and not with a staged rollout calendar. According to Cisco’s official blog, the system, called MyAgent, is live across the entire workforce and is designed to orchestrate cross-system work on behalf of each individual employee under supervised, governed conditions.
The Wall Street Journal reported that Thimaya Subaiya, Cisco’s EVP of Operations, described the rollout as one of the first examples of a widespread agent deployment inside a large enterprise, where autonomous bots are set to accomplish a variety of tasks on behalf of their humans. That framing matters: this is not a productivity add-on. It is a redesign of how a Fortune 500 company does work.
What MyAgent Actually Does
The core mechanic is delegation. An employee defines an objective. MyAgent determines what steps are required, coordinates across connected applications including Outlook, Webex, Jira, and SharePoint, and advances the work in the background under human supervision.
What makes this distinct from a chatbot integration is persistent memory. MyAgent retains each user’s preferences, past interactions, and work context across sessions. A follow-up task can reference what happened last week without the employee re-explaining the situation. The system builds continuity across work, not just access to several enterprise apps from a single interface.
Underneath MyAgent sits a much larger system. The 36kr report on the deployment details that MyAgent connects to more than 800 specialized agents in the background. Individual employees are not interacting with a monolithic model. They are directing a dispatcher that routes requests to purpose-built agents for specific tasks: email management, project tracking, market research, cross-system notifications, and more.
The Circuit Architecture and Why the Economics Work
The most operationally significant part of the Cisco announcement is not the agent experience itself. It is the cost architecture underneath it.
Running 90,000 always-on agents against frontier models would be economically unworkable. Cisco’s answer is the Circuit platform, which the company has been building internally since 2023 with a core team of roughly 40 engineers.
Circuit operates as a governed routing layer between employees and AI resources. According to the WSJ, the routing logic works approximately as follows:
| Request Type | Handling Layer | Share of Volume |
|---|---|---|
| Routine queries, structured tasks | Open-weight models on Cisco GPU clusters | 50-60% |
| Deterministic workflows | Traditional software automation and APIs | 20-30% |
| Complex reasoning, ambiguous requests | External frontier model APIs | Small fraction |
This tiered approach is what makes the economics survivable. Rather than routing every employee request to a frontier model at full API cost, Circuit assigns each task to the cheapest capable resource. Splunk, which Cisco acquired, provides token-level observability so engineers can monitor costs at the individual workflow level before they compound.
Cisco is also running open-weight models on its own data center GPUs, giving it marginal cost control that API-dependent architectures cannot achieve. The result is an agentic system that can operate at workforce scale without creating the kind of compute bill that would override any productivity gain.
From Chat to Ambient Intelligence
Cisco is explicit that MyAgent is not a chat interface. The company uses the phrase “ambient intelligence” to describe the intended mode of operation: the agent is always present, continuously aware of context, and available to take action across connected systems without the employee actively opening a tool.
That framing has direct implications for enterprise AI strategy. Most enterprise AI deployments of the past two years have been interactive, requiring an employee to open a tool, formulate a query, and interpret an output. MyAgent is designed to invert that model: the system knows what you are working on, remembers what you told it previously, and acts across systems while you focus elsewhere.
Subaiya described it in a LinkedIn post as a shift from passive, scattered AI experiments to an active Agentic Ops model, noting that Cisco is no longer using AI to improve individual steps in existing processes, but redesigning entire workflows around intelligent agents.
That is a meaningful distinction for enterprise leaders evaluating AI roadmaps. As the Salesforce Agentic Enterprise Index found earlier this month, most enterprise agentic AI deployments are still task-scoped automations rather than workforce-redesign initiatives. Cisco’s deployment is the latter category: a structural reorientation of how an entire company executes work.
The Governance Model Enterprises Are Watching
The question Cisco’s deployment raises is not whether the technology works. It is whether the governance holds at 90,000 agents operating simultaneously across enterprise data and applications.
Circuit’s access model addresses this directly. According to Cisco’s blog post, MyAgent operates within a governed environment where it only accesses approved models, approved systems, and enterprise-appropriate data pathways. Security and human oversight are built into the design from the start.
The practical mechanism is the routing layer itself. Because every agent action flows through Circuit before reaching a model or enterprise system, the platform can enforce access controls, audit every action, and apply cost limits at the task level. This is similar to the architecture that Snowflake’s Cortex AI Gateway introduced for multi-model agent governance: a control plane that mediates between agent requests and enterprise resources.
The Splunk observability layer adds a second control dimension. Token consumption, system access patterns, and workflow completion rates are all visible at the enterprise level. If an agent workflow starts consuming resources unexpectedly or accessing systems outside its expected scope, the monitoring layer can flag or halt it before costs or risks compound.
Why This Matters for Enterprise AI Teams
Cisco’s deployment establishes several reference points that enterprise AI teams should pay attention to:
Scale is achievable without frontier-model dependency. The 50-60% open-weight routing figure is significant. Enterprises that assume large-scale agent deployment requires heavy frontier model usage are overestimating the cost. A well-designed routing layer pushes the majority of volume to cheaper, faster, controlled resources.
Observability is not optional. Cisco’s use of Splunk for token monitoring is a first-class architectural decision, not an afterthought. At workforce scale, visibility into what agents are doing and what they are costing is what makes the deployment sustainable.
The blueprint model matters. Cisco is positioning the 90,000-employee deployment as a customer reference implementation. Enterprises evaluating Cisco’s AI product portfolio will now be looking at an internal production system rather than a demo environment. That changes the sales conversation.
Persistent memory is the differentiator. The technical capability most likely to change employee behavior is not the ability to delegate tasks. It is the agent’s memory of past work. Continuity across interactions eliminates the re-briefing overhead that makes current AI tools friction-heavy for repeated workflows.
For enterprise teams building their own agent deployment strategies, the Circuit architecture is worth studying closely. The combination of governed routing, open-weight cost reduction, observability tooling, and a workforce-level rollout is the closest thing to a validated enterprise agent blueprint that exists in production today.
Cisco’s AI orders growing from $2 billion in FY2025 to a guidance of $9 billion for FY2026 suggest the commercial momentum behind this bet is real. The 90,000-employee deployment is both a productivity initiative and a product proof point. For customers trying to justify their own agentic AI investments, Cisco is now the case study.
If you are planning an enterprise agent rollout, book a call with Enera to map your governance and routing architecture before scale reveals the gaps.