Extreme Networks Inc. used its Extreme Connect 2026 user conference this week to make a strong case that artificial intelligence-driven networking has finally arrived.

Building on Platform ONE, the company rolled out a full-stack vision spanning new Wi-Fi 7 access points, enhanced fabric-ready switching and a second-generation artificial intelligence layer called Agent ONE, designed to act less like a chatbot and more like an operational co-worker for NetOps teams. Framed by the insistence of Nabil Bukhari (pictured), chief technology officer and president of AI platforms, that Extreme is now an “outcome company” — and backed by several consecutive quarters of double-digit revenue growth — the announcements position Extreme well in the race to bring agentic AI to production networks.

What Extreme announced at Connect

The positioning of Extreme as an outcome company is enabled by Extreme’s “stack,” which starts at the hardware layer and extends up to AI to help customers meet business goals and cut operational costs. The evolution of that stack showed up in three buckets:

Extreme Platform ONE, expanded

Platform ONE, generally available since July 2025, is now the anchoring control plane for wired, Wi-Fi, fabric and software-defined wide-area network, with a strong focus on visualization and unified operations. This includes the following new capabilities:

• Single, “living” topology for physical, Wi-Fi and fabric layers, plus alerts/events, clients and inventory in one dashboard.
• Deep fabric visibility, so every node, link and service exposed, replacing the “fabric is a black box” complaint that kept coming up with customers and partners.
• Zero-touch provisioning and intent profiles for fabric, so teams can define once and push everywhere; Platform ONE monitors configuration drift and lets you snap devices back into compliance with a click.
• Integrated guest, location and wireless intrusion prevention system or WIPS services directly in the Platform ONE UI, eliminating separate portals and duplicate maps.
• “Edge Services” so third-party switches and access points (Cisco, Aruba, HPE, Juniper and the like) can be discovered and managed via a local on-premises service but visualized and orchestrated from Platform ONE.

Full Wi‑Fi 7 and switching refresh

Scott Calzia, vice president of product management, highlighted that Extreme was out of the gate early with Wi‑Fi 7 and is using Connect 2026 to showcase a broadening of its lineup for both high‑density and cost‑sensitive deployments.

• New WiFi 7 APs: 5022 and 5060 (indoor/outdoor 4×4 tri-band with dedicated sensor radio), plus 3020/3060 2×2 tri-band for entry-level and the 3020W wallplate for hospitality/dorm and retail use cases.
• On the wired side, new 5420M mixed-media switch (48port fiber/copper) and new 100G/400G options for the 7830, plus a universal ruggedized 4600 series to push fabric into operational technology/industrial spaces.

Agent ONE: Second-generation AI stack

When talking about AI, Bukhari was explicit that Extreme isn’t just “bolting a chatbot onto a frontier model.” He walked through a layered AI architecture:

• Frontier models and AI infrastructure at the base (sourced from hyperscalers, not built by Extreme).
• An Extreme AI Core is a networking-specific knowledge graph that encodes how MACs, clients, policies, sites and services relate across the Extreme universe.
• A skills layer where connectors, data pipelines and workflows live, so AI agents follow your standard operating procedures, change controls and tie into security, storage and compute, not just networking.
• An agentic layer, where Extreme Agent ONE operates in modes like Agent ONE Coworker (interactive copilot) and, later, Agent ONE Operator (autonomous operator).

The first mode, Agent ONE Coworker, is a proactive “warm coworker” that continuously monitors the network, investigates anomalies, and can execute changes on your behalf once you approve a plan. Extreme’s roadmap calls for a second mode, Agent ONE Operator, in Q4 CY26, to run workflows autonomously within guardrails.

I thought Bukhari summed up Extreme’s mission nicely when he stated, “Everything that we have thought of, everything that we have built, everything that we announce, is for one purpose, and that purpose is to make your life easier.” Too often, information technology vendors create new solutions but forget about the practitioner. Network operations have grown increasingly complex, and one of the first orders of business for AI should be to let engineers do what they need to do, but much more easily and quickly.

Customer traction on display

Extreme sprinkled real-world stories throughout the keynotes to demonstrate the value of Platform ONE and the new hardware. Chet Patel, director of innovation and technology for Caribe Royale Orlando, which hosted Extreme Connect, discussed how Extreme helped reduce Wi-Fi-related complaints. “We had hundreds of tickets before we deployed Extreme, and then after, we now have zero tickets,” he said. “It’s now ‘set it and forget it,’ which is not something we could have ever said about Wi-Fi before.”

In the “drinking your own champagne” portion of the keynote, Extreme’s own Chief Information Officer Anisha Vaswani explained the value of cross-portfolio integration. “We used to have one tool for managing on-prem, we had XIQ for wireless, and we had Ipanema (acquired by Extreme) for SD-WAN,” she said. “Today, we can manage all of that in one platform, and it’s made our lives easier and streamlined operations.”

Another customer, Richard Gingerich, systems engineer for Sight and Sound Theatres, talked about the value of Extreme’s fabric and the desire for an agentic style interface to assist the helpdesk. “I would love for the help desk to interact with the AI agents, he said. “Just ask the bot, ‘Bob has a problem, what’s going on?’ or ‘What VLAN is port 48 on?’ without having to come to the platform UI.”

Those quotes reinforce that Extreme’s AI story isn’t only front-of-house. They’re designing agents to serve as front ends for nonexpert tiers of the organization.

Financial performance underscores transformation success

It has been a long journey for Extreme. When Chief Executive Ed Meyercord (pictured below) and Norman Rice joined Extreme, the company’s viability was in question. A number of strategic acquisitions and a few years of engineering work have put the company in a strong position, with a simplified, strong enterprise portfolio.

Onstage, Chief Marketing Officer Monica Kumar and Bukhari both referred to “feel the momentum,” which is supported by recent numbers. Extreme has quietly put together a multiquarter run of solid growth while shifting the business mix to SaaS. The most recent quarter, the third of fiscal year 2026, saw revenues of $316.9 million, up 11% year-over-year, marking the fifth straight quarter of double-digit growth. Software-as-a-service annual recurring revenue is now $236.4 million, up 29%.

On the market side, Extreme’s stock is trading in the mid-20, up 42% this year and almost 70% in the past 12 months. Extreme Connect hosted several investors, who appear bullish on the company’s outlook.

Takeaways for network practitioners

Though Extreme Connect addresses many audiences, including mine, the most important audience is the network practitioner – the people who work day after day to keep networks up and running, ensuring businesses function. If you’re running a campus, distributed enterprise or OT-heavy environment, there are three practical angles from Extreme Connect that matter:

AI that understands networks, not just text

The Agent ONE design, which includes a knowledge graph, skills and an agentic layer, is an attempt to encode networking domain expertise so AI can do more than summarize PDFs. That shows up in tangible capabilities like real-time five-second packet streaming per client for Wi-Fi troubleshooting in Platform ONE, AI-driven control via conversation, and nudges that surface anomalies before users complain.

Operational convergence is becoming table stakes

Extreme is betting that “one pane of glass” is finally real, not just marketing: wired, Wi-Fi, SD-WAN, fabric, guest, WIPS and even multivendor assets hanging off “edge services” all land in Platform ONE. For practitioners, that means you can reasonably push vendors to show how their AI and observability work across the full topology, not just their own APs or switches.

Design for human-in-the-loop governance

Bukhari’s talk stressed that AI should optimize joint performance of human and machine,” and that humans must remain in and on the loop. As you evaluate AI-heavy networking offerings, including Extreme’s, the questions to ask are:

• Can I see every action the agent proposes and approve/deny it?
• Can I encode my change windows, rollback policies and site reliability engineering practices as skills?
• How easy is it for Tier1 to safely interact with the agent without blowing up the network?

Most IT leaders I speak with are bullish on AI as an IT tool, but many engineers remain cautious because you often don’t know what you don’t know until it’s too late. Looking ahead, here are a few pragmatic steps to move forward, while minimizing risk.

• Start piloting AI-assisted operations in low-risk domains (Wi-Fi troubleshooting, reporting, configuration validation) and capture hard metrics on MTTR and ticket volume.
• Push vendors to expose their “AI core” – including data models, guardrails and integration points – rather than accepting black-box copilots.
• Use these agentic platforms to bring networking closer to identity, security and ITSM. Extreme’s integrations with Entra ID, security information and event management and ServiceNow are a good benchmark to use.

Final thoughts

The broader story emerging from shows like Extreme Connect is that networking is finally entering the same AI-first transition that has already reshaped software development and security operations. As agentic systems and domain-specific copilots move from demos into everyday workflows, expectations for NetOps will shift. Platforms will be judged less on how many knobs they expose and more on how effectively they translate intent, policy and telemetry into closed-loop, measurable outcomes.

Zeus Kerravala is a principal analyst at ZK Research, a division of Kerravala Consulting. He wrote this article for SiliconANGLE. 

Photos: Zeus Kerravala