The Operating System for Enterprise Networking

The first generation of Extreme AI ran across thousands of enterprise networks. We watched what engineers asked it, where it helped, and where it fell short. More importantly we talked to the practitioners—and we listened.

You wanted it to be proactive. Ambient. To come to you rather than wait to be asked. So we built Extreme Agent ONE™ Coworker. We introduced that yesterday.

But you had more to say. You wanted it to keep working when you are not around. To run your disciplines without being reminded. To handle what it could while you focused on more important things—or caught some sleep, if the network allowed it.

That is why we are here today.

The Second Relationship

Extreme Agent ONE™ Operator works for you whether you are present or not. It is autonomous, always-on, and extensible. It acts within the governance boundaries you set. It remembers everything it learns about your network and your team. And it keeps getting smarter from every task it runs, every interaction it has, and every skill added to its repertoire.

That is not a feature difference. That is a fundamentally different relationship with AI.

Meet Agent ONE Operator

When designing Agent ONE Operator, we wanted it to feel familiar. So we modelled it after you. The best way to understand Agent ONE Operator is to think about how you work.

You act when new information demands your attention. You act because certain things are part of your discipline. And sometimes, you act obsessively because the stakes demand it.

The operator works the same way. It performs tasks.

When something on your network demands attention, it runs a task. And if you have given it permission to act, it already has. It does not wait for you to come online. It does not draft a recommendation and hold it until you return. It acts within the boundaries you set and has a summary ready for you when you are back. The operator does not interrupt your Sunday. It handles it.

It runs your routine tasks on whatever schedule you define. The morning summary you want before your first meeting. The weekly trend report you review every Monday. The nightly check on your most critical sites. Described in plain language. Executed without you being present. The operator maintains your disciplines even when life gets in the way.

And for the tasks that demand constant attention, it runs at whatever frequency you need. Daily. Every five minutes. Continuously. You set the frequency. The operator holds it. Without fatigue. Without distraction. Without ever deciding it has checked enough times.

One concept. Tasks. Triggered by events, scheduled by you, or running continuously. Three ways of working that mirror exactly how you work.

Now think about how you remember.

You have short-term focus. What you are working on right now, in this conversation, in this task. When it ends, that focus moves on. You have experiential memory. What happened over time. What worked, what failed, what you learned. The accumulated log of your experience.

You carry institutional knowledge. What normal looks like in your environment. Your organization's risk tolerance. What your team cares about. The things a great engineer knows that are not in any runbook. And you have muscle memory. How things get done in your specific environment. The patterns, the escalation paths, the approaches that work here. You do not think about them consciously anymore. You just know.

The operator carries all four. Modelled after human memory. Because it works for you. All four compounding over time. Making the operator smarter, more precise, more yours the longer it works with you. It lives everywhere you work. Inside Extreme Platform ONE™. In a standalone app. In Microsoft Teams. In Slack. The operator reaches you wherever work happens. You do not need to open a dashboard to stay in control. The operator comes to you.

We modelled it after you. Because it works for you.

First Contact

The first time you meet Agent ONE™ Operator, it does not ask you about your network. It asks about you. Are you an engineer who prefers technical detail? Or an operations leader who wants trends and implications? The operator learns your communication style and from that moment every interaction is calibrated to you.

Then it asks how you want it to operate. What it is always authorized to do. What it must always ask you before doing. What the boundaries are. You answer in plain language. No forms. No configuration panels. A conversation. The operator configures its own governance structure and comes online.

And then it goes to work. No lag. No preamble. Whatever your network needs at that moment, the operator is already on it.

Straight to work. Straight to value.

How We Extended the Stack

Agent ONE Operator required us to extend the Extreme AI Stack. And so we did.

The Agentic Layer

The agentic layer had to evolve from a system that executes when called to a system that operates continuously whether called or not. That required three things:

• First, a four-type memory architecture that mirrors human memory. Working memory for the active session. Episodic memory for the accumulated log of interactions and outcomes. Semantic memory for what the operator has learned to be true about your network and team. Procedural memory for how it operates in your specific environment. Each type required different storage, different retrieval, different reinforcement models.
• Second, a robust scheduling engine capable of managing tasks of varying frequency, priority, and complexity simultaneously, all running continuously without user presence.
• Third, the harness, extended significantly for Operator. Agent identity management: every action is traceable to a verified identity. Full auditability and observability—every decision, every action, every outcome logged and explainable. And deep native connections to Extreme Platform ONE's common services, not an API integration but an architecture-level connection. They share the same data model. They speak the same language. When Extreme Platform ONE knows something, the operator knows it instantly.

The Skills Layer

In my previous blog post, Introducing Extreme Agent ONE—Built Different, the Skills layer encoded the Standard Operating Procedures (SOPs) of networking as executable intelligence. For Operator that foundation becomes extensible.

The operator needs to keep getting smarter. Not just from Extreme, but from everyone who understands enterprise networking. That is why we built Extreme Exchange™, an AI skills catalogue, community, and lifecycle management layer where skills developed by Extreme, partners, and customers are discovered, activated, and managed in one place. Every skill added makes the operator more capable.

If you build solutions for enterprise networks—security, ITSM, analytics, vertical platforms—Extreme Exchange is where your intelligence meets ours. These conversations started this week. Operator was built for the future no one can fully see. Extensible skills, discovered, activated, and managed through Extreme Exchange is how we get there.

AI Core and AI Infrastructure

AI Core expanded to support the deeper memory model, richer governance, and more sophisticated context that always-on operation requires. AI Infrastructure keeps advancing on its own. Every new frontier model makes the operator more capable automatically. The foundation rises. The operator rises with it.

NVIDIA President and CEO Jensen Huang said OpenClaw is the operating system for personal AI. This is our equivalent moment for enterprise networking.

Agent ONE Operator is the operating system for enterprise networking in the agentic era.

Not a product feature. Not an AI add-on. The operating system. The layer through which every capability of the Extreme AI Stack comes to your network team as an intelligent, autonomous, always-on operator.

The era of bolt-on AI is over.

The Close

Let me end where it matters most.

Agent ONE Operator will save your Sunday. The Monday morning summary will be waiting. But we know this is just the beginning.

As you use Agent ONE Operator you will want more. You will find new ways to use it. You will bring us challenges we have not anticipated yet. That is the point. The stack we built will expand because your needs will expand.

Building the Extreme AI Stack was hard. The architecture, the technology, the execution—that kind of hard we know. The new hard is different. It is building trust between you and an AI that acts on your behalf. Every governance boundary we set, every permission model we designed, every audit trail we built exists to answer one question: how do we make it possible for you to trust Agent ONE Operator with something that actually matters. That challenge we have never faced before. And it is the one we are most committed to getting right—together.

Nabil Bukhari

President of AI Platforms and EVP, Chief Technology Officer, Extreme Networks. Responsible for all product, all engineering, and the P&L for the entire recurring business.