Collaboration events have rarely been must-attend shows for network engineers. Meetings, contact centers, and room systems ran on the network, and the network team kept latency low and Wi-Fi up. At Cisco’s WebexOne 2026 event in Austin, it became clear that this division of labor is ending.
Over two days of keynotes and analyst sessions, Cisco President and Chief Product Officer Jeetu Patel and his team described AI agents that join meetings, work around the clock, and communicate with each other at machine speed, as well as room devices that function as sensors. All of that lands on the network. Here are the themes that should be on the radars of network engineers.
Agents create a new traffic profile
The most important data point for network teams came from Patel’s day-one keynote. Chatbot traffic, he said, was “very human-bound,” with spiky usage patterns that mirrored people’s work habits. Agents are different. “These agents work seven by 24,” he said, and they spin up subagents that do the same.
Patel followed up by stating that agent token consumption surpassed human consumption in February and is now five times higher, citing OpenRouter data. He also said that an agent performing a task consumes 450% more network bandwidth than a human doing the same work, and that 60% of global compute this year will go to inference rather than training.
The day-two demos showed what that looks like in practice. In a customer service scenario built on Cisco’s new Dialog platform, Jessica Smith, director, product marketing, noted that the agents had “communicated agent to agent at machine speed” to arrange a vacation package while the customer went about his day.
That means assumptions about peak hours, oversubscription, and branch uplink sizing need another look. Sustained, machine-generated traffic doesn’t follow the patterns most campus and WAN designs were built around.
The room is now a network endpoint and a sensor
Espen Løberg, Cisco’s vice president and general manager for devices, described collaboration devices in terms network professionals will recognize. “They are the last mile of secure AI infrastructure in the workplace,” he said, adding that they can “see, hear and sense what’s happening around them with AI running directly on the device.”
Cisco’s new Ceiling Speaker Pro is a good example. It integrates Cisco’s ceiling microphones, cameras and codecs into a fully IP-connected room, replacing multi-vendor audio setups with a single Ethernet cable. Snorre Kjesbu, senior vice president and general manager of Cisco Collaboration, explained the rationale: “IP gives you power over IP, it gives you media over IP, and it gives you management and manageability over IP.” That means PoE budgets, switch port planning, and media QoS now include gear that used to be part of a separate AV system.
Workplace Designer, which now spans entire buildings rather than single rooms, ties everything together. In the demo, it recommended access point counts and placement alongside room devices. As the Cisco product manager running the demo noted, “Video collaboration requires more access points, and use cases like asset tracking and wayfinding require even greater density.” Wi-Fi design and room design are now the same exercise.
One management plane changes who owns what
During the analyst Q&A, Patel was very matter of fact about Cisco’s platform consolidation. Cisco Cloud Control, which he said came together in nine months, is now the single management plane for Cisco products, and Webex Control Hub has been folded into it. “Over time, Meraki as a brand will be gone. Over time, Catalyst as a brand will be gone,” he said. (Read more: What is Cisco Cloud Control and why should customers care?)
Collaboration, networking, security, and observability are now managed side by side. Intelligent Workplace Experiences ties network and device assurance to floor plans, Wi-Fi coverage, and occupancy, enabling IT to troubleshoot by business impact. Workspace Advisor, Løberg said, “uses AI inferencing to create a real-time digital twin” of rooms to spot issues before users report them.
Anurag Dhingra, who leads Cisco’s intelligent workplace efforts, described the operating model behind it during the analyst session: ambient agents that monitor infrastructure, reason about root cause, test fixes in a digital twin, and then deploy them, with an “autonomy dial” that lets operators decide how much to delegate. It’s the same agentic ops model Cisco uses for networking, now applied to rooms and floors. The boundaries between network, collaboration, and facilities teams are blurring, and someone will need to own the combined workflow.
Security moves to runtime, and agents need identities
Patel argued that security and observability, historically separate markets, are converging. “You cannot enforce security through static policy at design time,” he told analysts. “They’re dynamic policies enforced at runtime.” This theme was prominent at the recent Splunk .conf and Cisco Live events.
Andrew Cohen of Anthropic’s applied AI team shared a practical checklist during his keynote portion. What you ask of an agent, he said, is what you’d ask of an employee: respect data boundaries, ensure “clear identification, authentication,” and make all work observable. Claude agents in Webex have scoped permissions and agent tracing, so every action can be reviewed.
Cisco’s AI Agent 360 adds controls for Webex AI agents. In the Dialog demo, an agent was blocked from approving a refund it wasn’t authorized to grant, and Cisco AI Defense redacted personal health details the customer had volunteered before they reached downstream systems. Splunk agent observability then flagged a recurring issue and recommended a fix. For network and security engineers, agents are becoming another class of identity on the network, and their telemetry will sit alongside flow data and logs.
Hybrid inference is the new hybrid cloud
Patel also argued that “control is the new moat.” Regulatory, sovereignty, and intellectual property concerns mean some inference will run locally. That theme appeared on day two as well. AWS described its partnership with Cisco on an LLM Proxy that adds guardrails while allowing builders to choose models via Amazon Bedrock. Vinod Muthukrishnan, who runs Cisco’s customer experience business, said Cisco is “doubling our investments” in on-premises contact center platforms. Expect AI workloads to be split across the cloud, data center, and edge, with the network responsible for secure, predictable paths between them. Webex Contact Center has many public-cloud competitors, as well as a couple that deliver solutions only on-premises, but Cisco is the only vendor that can deliver a true hybrid solution.
Cisco’s integration advantage
I’ve been to every WebexOne since the event began, and this was the best showcase of Cisco’s “end-to-end” platform. Much integration has occurred since Patel became Chief Product Officer and broke down historical product silos. Kjesbu often talks about how Webex “stands on the shoulders of giants,” meaning the Webex products are built on Cisco’s market-leading security and networking products, but the value is actually two-way. Cisco devices and software clients hold telemetry that network and IT operations can use to optimize the user and customer experience. (Read more: How Jeetu Patel made Cisco unrecognizable)
The collaboration pure-plays all have strong UCaaS/CCaaS offerings but no networking or security, while the network vendors lack the collaboration stack. Cisco has both, along with Splunk for observability and AI Defense for security, all managed through one platform. The same Cloud Control instance that detects a struggling access point can connect it to a choppy Webex call or a failing room device, and Splunk can correlate telemetry, including from contact center agents.
Patel highlighted the benefit during the analyst Q&A: “Every subsequent technology you buy should cost you less to integrate than the previous one,” and each purchase should add value to everything else you already own. The advantage holds only if integration delivers and there is no lock-in, which is why native Google Meet, Teams, and Zoom support on Cisco devices matters. For network engineers, the integrated stack is a chance to move from keeping the pipes running to owning the experience end to end.
In the past, the “Webex on Cisco” advantage was nominal and a bit of a solution looking for a problem, but in today’s AI-first world, the problem is ensuring agents have access to the right data, perform well, and stay secure, and now the full stack matters.
Final thoughts
Patel explained that Cisco’s campus and branch business, historically tied to GDP growth, has grown more than 20% over the past couple of quarters as companies re-equip offices. He also cited 1.4 billion square feet of commercial real estate that’s up for lease renewal. That’s a large refresh cycle, with the network at its center. WebexOne made clear that in the agentic era, collaboration is no longer an application sitting on top of the network. It’s one of the biggest drivers of what the network must do.
For network engineers, I’d suggest five steps:
- Baseline agent traffic now, especially for teams adopting coding and personal agents, and revisit uplink and WAN sizing assumptions.
- Get involved early in workplace and room design; access point density, PoE, and media QoS decisions are being made in tools that also place cameras and speakers.
- Push for agent identities to be managed like any other non-human identity, with ownership, scoped access, and revocation.
- Connect network telemetry to the observability and security platforms that will monitor agents, because that’s where problems will surface first.
- Map where inference will run, whether in the cloud, data center, or at the edge, and design for those paths before the workloads arrive.
Network engineers have spent years being told the network is strategic. With agents, it finally is, and the teams that get ahead of this shift will shape how their companies work, not just connect them.