Introducing Extreme Agent ONE™—Built Different

AI is here, whether you invited it or not!

AI believer or AI skeptic. AI waits for no one. It has already operationalized in your enterprise. It is in the hands of your users, in the applications they run, the tools they use, the workflows they depend on every day. It does not wait for organizational consensus. It does not ask for your permission. And those AI applications are fast. They provision in seconds. They respond in seconds. They expect the network beneath them to keep up. So now what?

Three forces are driving what comes next. AI demands a network managed at machine speed. Identifying issues, troubleshooting, remediating. All faster than any human team can consistently deliver. The engineers to manage tomorrow's networks manually simply cannot be hired fast enough. Or in many cases, cannot even be found. And in every industry, the organizations already moving on this are not AI enthusiasts. They are pragmatists who looked at the math and realized it doesn't work any other way.

Current generation AI helped. It made humans faster. We know—we ran it across thousands of enterprise networks and watched what engineers actually needed from it. But faster humans are still humans. And that is no longer enough. This is why an update wasn't going to work. Evolution wasn't enough.

So we set out to build an entirely new generation of AI. One that delivers on three promises.

Today I am introducing the next generation of AI at Extreme.

Promise One—The Right Technology

The first promise is the right technology. And technology at this level does not come from a single product bolted onto a frontier model. Anyone can do that. Ask a generic AI about networking and it will give you an answer. It might even be a good one. But it will not know your network. It will not know that site 14 has been running hot for three weeks. It will not know your change window is Sunday nights or that the firmware on your access points has a CVE that affects your specific topology.

Context is what separates genuinely useful AI from generically capable AI. And context is built in layers.

NVIDIA President and CEO Jensen Huang showed the world what that looks like at the infrastructure level. Five layers, all advancing simultaneously. Today I am showing you what it looks like for enterprise networking. The Extreme AI Stack.

Four layers. Each one building on the one below.

AI Infrastructure—Reasoning

The foundation of the stack is reasoning capability. The world's best frontier models from the leading cloud providers and AI labs bring the latest and greatest reasoning capability to everything above them. This is not a static foundation. As new models arrive, as reasoning capabilities advance, as the frontier moves, everything built on top of it benefits automatically. Our customers inherit every breakthrough in global AI without lifting a finger.

But raw reasoning alone is not enough. Reasoning about what?

AI Core—Context

This is where Extreme's deepest competitive advantage lives. AI Core answers the question the foundation leaves open. It gives the reasoning capability something specific, rich, and uniquely Extreme to reason about. Your network.

AI Core provides three things no generic AI can replicate:

• Real-time and historical access to your network data. What is happening right now and what normal has looked like over time.
• Domain knowledge. AI Core does not just receive data from your network. It understands what that data means. Every entity in the Extreme world is defined, understood, and contextualized.
• A knowledge graph that understands the relationships between every entity. It knows that a MAC address connects to a user identity, that identity connects to an access policy, that policy connects to a site. And it keeps going. Every relationship, every dependency, every edge case that makes real world enterprise networks as complex as they actually are. Not through configuration. Inherently.

Combined with the governance framework that defines what the AI can act on autonomously and what it brings to a human, AI Core is the intelligence engine that makes everything above it possible. This is the moat. Not the model. The context, the domain knowledge, and the knowledge graph the model is born into.

Now you are reasoning in context. But do you know how?

The Skills Layer—Procedural Knowledge

The Skills layer answers that question. It brings three dimensions to the stack:

• Tool access: the ability for agents to reach out and interact with systems, platforms, and data sources across your entire technology environment.
• Data connectivity: bringing context from other ecosystems into the intelligence layer. Identity systems, security platforms, cloud environments. The network does not exist in isolation and neither does the AI managing it.
• Procedural knowledge: the third dimension, the one that is hardest to replicate and most consequential. The SOPs of networking encoded as skills the agents can execute.

A frontier model can reason its way around how to configure a network or troubleshoot a fault in your environment. But it does not know how to do either with your safeguards, your change management window, your specific operational constraints.

It can walk you through a generic network design, but it does not know the precise procedure for designing a world class venue network the way Extreme does. Where we are the market leaders. Where the steps and sequence that separate good from exceptional have been perfected across hundreds of deployments.

In every case the frontier model has the knowledge. What it does not have is the procedure. The Skills layer brings that procedural intelligence into the stack. Not as documentation but as executable intelligence the agents can act on.

Now you are reasoning in context with the knowledge of how to do everything. There is one question left. Who actually does it?

The Agentic Layer—Execution

The agentic layer is where everything above it comes to life. It is the full technical infrastructure required to execute at enterprise scale. Reliably, securely, continuously, in a way a real enterprise can trust.

What you experience as one intelligent entity working on your behalf is an orchestrated system of specialized agents working in parallel. A supervisor agent decomposes the task and coordinates the work. Specialist agents execute within their domain. Network analysis, security posture, configuration validation. QA agents validate before anything reaches you. And wrapping all of it is the harness. The enterprise-grade infrastructure that manages agent lifecycle, enforces governance at every step, maintains persistent memory, and produces an auditable record of everything that happened and why. Without the harness you have a capable demo. With it you have something an enterprise can rely on.

Reasoning. Context. Procedural knowledge. Execution. Four layers. One coherent stack built specifically for enterprise networking.

All that complexity. Invisible to you. What you see is a network that finally keeps pace with the AI running on top of it. Reasoning at the speed of frontier models. Context drawn from every corner of your environment. Procedural knowledge perfected across decades of enterprise networking. Execution at machine speed, around the clock. That is the first promise. The technology your network demands. Finally delivered.

Promise Two—The Relationship

The second promise is about you!

Not your network. Not your infrastructure. You. The person. The network engineer who has not slept through a Monday night in years. The operations leader managing a team stretched thin across a network that grows more complex every quarter. The practitioner who knows their environment better than any AI ever will, and who deserves a technology that works with them rather than demanding they work around it.

We built the stack for your network. But for you, we built Agent ONE.

Agent ONE is how you experience the entire Extreme AI Stack. Every layer. The reasoning, the context, the procedural knowledge, the execution. All of it comes to you through a single entity. Not a dashboard. Not a chatbot. Not another tool. A coworker.

When we were building Agent ONE, I pushed the team hard on one question: what does it actually mean to be a coworker? A coworker is helpful. Specifically helpful in the context of what you are doing right now. A coworker is smart. They have already done the work before you ask. And a coworker has a personality. They speak up when something matters. They notice when you are stuck and offer a hand. They do not wait to be invited into the conversation.

So we gave Agent ONE a personality. A real one. Not robotic. Not scripted. Someone you actually want working alongside you.

And the way Agent ONE expresses that personality, the way it reaches out, the way it speaks up when it matters. We gave that a name.

We call it the nudge.

Agent ONE does not wait to be asked. It notices. It reaches out.

A Sev 1 alert fires at 2am. Before Agent ONE reaches out it has already investigated. It does not wake you up with an alert. It wakes you up with findings. And it will be persistent. Because a real coworker does not let something critical go unnoticed.

You have been on a configuration page for a while. Agent ONE notices. It has something relevant to say. It surfaces it gently, non-blocking, at the edge of your view. Informational. Not demanding your attention. Just offering it.

Like a real coworker, Agent ONE reads the room. The urgency of the nudge matches the urgency of the moment. Critical. It tries harder. Informational. It stays gentle. Never annoying. Always relevant. Always at the right moment.

Your team will start talking about it this way. "Hey, did you see the latest nudge?"

That is exactly what we designed for. Not a feature. A word that enters the vocabulary of your operations team and stays there.

The hardest part, honestly, was stopping it from being annoying.

Promise Three—It's Now

The third promise is the most important one. Because in this industry promises are easy. Visions are announced at conferences and arrive two years later. If they arrive at all. Roadmaps are dressed up as reality. Futures are sold before they are built.

We did not want to do that. We wanted this to be now.

And so it is.

Agent ONE Coworker ships with six skills. Five you already know from the first generation, improved, sharper, faster, now running on the full Extreme AI Stack. Talk to Data: your network answers in plain English. Talk to Knowledge: GTAC, product documentation, and your specific CVE exposure, all in one conversation. Talk to Support: open, track, and resolve support cases without leaving Agent ONE. Canvas: the dashboard you need right now, built from live data on demand, not from a template someone designed six months ago.

And two that are new in this generation. Talk to RRM: Radio Resource Management via conversation. Govern and optimize your Wi-Fi environment through Agent ONE. This is the Skills layer doing exactly what it exists to do. Turning intelligence into action. And Nudge: the skill that gives Agent ONE its personality and its voice. The one that makes it a coworker rather than a tool.

Every Extreme Platform ONE customer gets upgraded to Agent ONE Coworker starting July 1st. You do not fill out a form. You do not schedule a migration. You do not do anything. You log in. Your coworker is there.

This is not a vision. This is not a roadmap. This is what you get. Now.

A Personal Note to Close

I want to close with something personal.

At Extreme we don't follow the market. We follow the future. We don't react. We anticipate.

Extreme Platform ONE was not a reaction to market demand. It was a bet on where complexity was heading. First generation AI was not a response to competitive pressure. It was a conviction that human attention alone would not be enough. Both took years to build and moments to validate.

The Extreme AI Stack is our next bet. Bigger stakes. Bolder outcome.

Today we delivered the future. But we're not done. Tomorrow, we take it further.

I will meet you again in part two.

Nabil Bukhari