Extreme Connect 26: Agent ONE takes forward network AI

With the company’s president of artificial intelligence (AI) platforms and chief technology officer (STO) Nabil Bukhari telling his customers that their businesses are not based on the networks that they signed up to just a short time ago, Extreme Networks has opened its annual Connect user conference with the launch of Agent ONE, described as a new class of AI agents for enterprise networking that moves beyond generic, prompt-based AI.

The launch came at a time when Extreme’s customers had clear expectations form its product line. Customer surveys were showing three main demands: 93% trusted AI-powered networking; 94% were ready to invest in networking AI technologies; 57% were expecting results from their products in weeks.

Bukhari emphasised the clear progression and evolution in networks over the past five to two years: “These networks we’re building and running are bigger, more complex and have more devices in it. Your companies rely on it a lot more, and your users want results faster. And what about your teams? They’re probably shrinking. These are not the networks that you signed up. 

“As networks begin to think, adapt and act in real-time, the relationship between human users and AI agents will rapidly evolve, making simplicity and control essential to success. Our vision is autonomous networking at scale delivered on a foundation of trust between humans and AI agents, which means fewer disruptions, faster outcomes and operational efficiency.”

To that end, Agent ONE runs on an AI stack that is said to be purpose-built for enterprise environments, which combines advanced AI reasoning, live network context and operational expertise to transform enterprise networks into systems that detect and act autonomously in the established governance framework. As a result, said the company, customers experience fewer disruptions, faster outcomes and networks that operate at the speed of the business.

Part of the Extreme Platform ONE system, Agent ONE is seen by Extreme as an entire new generation of AI that the company needed to release. The AI stack had four levels: an AI infrastructure layer, AI core, a skills layer and agentic layer. These would respectively provide reasoning, contest, procedure and execution, with a harness on top of the agentic layer to provide governance and adhere to businesses’’ own rules.

Bukhari said: “This is the stack, the four layers that work together. The infrastructure layer provides reasoning, the AI core provides context, skill layers provides the SOPs and the procedures, and then the agentic layer builds the execution on top of it. It’s complicated, but this is the stack that delivers outcomes through the AI.”

The initial part of the launch will be the introduction of Agent ONE Coworker in July 2026. Agent ONE Coworker is essentially an AI agent designed to work alongside IT teams and deliver proactive, context-aware intelligence with real-time decision-making and automated execution at machine speed.

Unlike traditional AI tools that wait for prompts, Agent ONE Coworker is attributed with being able to provide insights and guide decisions within the workflow. The company said that it had the core qualities of being ambient – it’s already done the work, proactive in that it does not wait for input, and is helpful – it understands and can do more.

Agent ONE Coworker is attributed with delivering conversational access to network data, documentation and security insights; automated support workflows from case creation through resolution; on-demand, real-time dashboards built from live data; and AI-driven Wi-Fi optimisation through conversational control.

With access through a single conversational interface, the module continuously monitors network activity, investigates anomalies and is designed to act in a way that it can reduce resolution times, minimise manual effort and prevent issues before they affect users.

A key feature of the platform is a “nudge” capability said to be able to deliver timely, contextual recommendations that turn insight into immediate action. Bukhari told the conference that in the first generation of Extreme AI networking tool, there was an AI expert management tool but it needed to be prompted, but that has now changed.

“In the second generation, we wanted to be the other way around, where you don’t have to go find the AI,” said Bukhari. “The AI finds you and prompts you. The second thing that we wanted to do was for it to be helpful ... and whenever it feels that it needs to talk to you, it prompts you. This is a relationship, so we wanted it to feel like a relationship. We wanted Agent ONE Coworker, to have a personality – to reach out [and] talk to you. So, we gave it a personality and we gave it a way to reach out.”

Extreme saw a potential key use case for Agent ONE Coworker in detecting rising Wi-Fi congestion and recommending or automatically applying a fix or identifying recurring point of sale (POS) slowdowns in retail to suggest traffic prioritisation during peak hours, turning patterns into immediate, low-effort decisions.

The next release and second mode of the system later in 2026 will be Agent ONE Operator, an always-on, autonomous agent designed to extend AI beyond real-time interaction to continuous network operation. It is also intended to execute tasks independently in defined governance boundaries, responding to events in real time and running scheduled workflows without requiring constant human input. It will continuously learn from each interaction and outcome, becoming more precise and effective over time.

Extreme said this evolution represents a shift from AI that assists in the moment to AI that operates continuously, ensuring networks are always monitored and optimised. Another key development is Extreme Exchange, an AI skills marketplace for Extreme Platform ONE that enables customers to discover, activate and manage skills that extend Agent ONE Operator’s capabilities.

It is built to deliver domain-specific intelligence across industries such as healthcare, education, retail and manufacturing, while integrating with IT service management, security, observability and cloud platforms. Built on an open model, it supports first-party and partner-developed skills and is designed to support customer-created skills in the future.

Assessing the new AI products and the impact that they may have on the industry, leading analyst Zeus Kerravala, founder of ZK Research, said: “Most vendors are still delivering AI as copilots. Extreme is taking a different path – embedding reasoning, context and execution into the network itself. That’s a meaningful step toward true autonomous infrastructure and a clear signal of where the industry is going.”