A phone cannot connect to the wireless network. An operator types a natural language prompt into Cisco Cloud Control. The platform dispatches a topology agent, routes through a security domain, crosses from networking into firewall configuration, identifies a missing Open Shortest Path First protocol route advertisement, and surfaces a one-click remediation. The sequence that would have consumed hours of ticket escalation across multiple teams resolves in a single conversation, without anyone logging into a second dashboard.

That demo, shown live during the Cisco Live 2026 keynote in Las Vegas, is the clearest illustration of what Cisco is building. I was in the analyst and press sessions across two days of keynotes, executive roundtables, and product briefings. The breadth of what was announced is significant, and the reason it holds together is not a product roadmap. It is what Cisco has built over forty years inside enterprise networks.

Chuck Robbins opened the keynote with a point that is easy to say and harder to earn: every major technology transition, from the internet to mobile to cloud, has run on the network, and Cisco has been the infrastructure layer for each of them. The agentic era is the next transition. The difference this time is that agents operate continuously, generate 450 percent more network traffic per task than a human performing the same task, and invert the upstream-to-downstream traffic ratio that current network architecture was designed around. I covered Cisco's WAN research on this inversion earlier this year. Cisco Live 2026 confirmed it with product announcements that treat it as an engineering constraint, not a marketing observation.

Jeetu Patel, President and Chief Product Officer, told the analyst group that two years ago he said Cisco would be unrecognizable as a company in two years. The announcements this week are his evidence. What makes that claim credible is not the number of announcements. It is that each one is built on telemetry, hardware, and operational data that Cisco has accumulated from running inside the largest enterprise networks on the planet. A software-only platform can describe the same capabilities. It cannot source them from the same place.

Cisco Cloud Control unifies Catalyst, Meraki, Nexus, WebEx, Intersight, and Splunk under a single login, a shared data layer, and a common agent framework. Humans and agents work from the same operational context. The platform runs three purpose-built models underneath: a Deep Network Model trained on decades of Cisco operational data, a Foundation Security Model, and a Time Series Model for infrastructure telemetry. The Studio feature lets customers build their own agents and applications in natural language directly within the platform.

Fifty-two partners were building inside the marketplace at launch, including ServiceNow, Slack, PagerDuty, and Snowflake. Patel noted that Cisco brought those 52 partners to the platform in a matter of weeks. Hasmukh Ranjan, CIO of AMD, joined Patel on stage as a customer voice for the platform.

Underpinning all of it is the Cisco Data Fabric, a common data platform powered by Splunk that ingests telemetry from networks, security systems, applications, and third-party sources at scale and correlates it into a single action layer. In a conversation at the Cisco Live evening reception, Kamal Hathi, Executive Vice President and General Manager of Data and Observability at Cisco, made a point that reframes who the platform is actually for: the Data Fabric is not only an infrastructure tool for technology teams. Business leaders can draw from the same correlated telemetry to understand operational performance, not just IT health. That expands who sits at the Cloud Control interface and what decisions they are making from it.

Cisco calls the operating model AgenticOps: humans and agents working from shared context, with humans retaining approval authority over the actions agents surface. The organizational implication is visible in how Starbucks CEO Brian Niccol described it from the keynote stage. Niccol is moving fast on AI across supply chain, store scheduling, and customer experience. Varadarajan owns the architecture that makes that speed defensible. Cloud Control is the platform that reduces the coordination cost between those two responsibilities, because an agent that crosses networking, security, and observability domains no longer requires three teams and two escalation queues to resolve.

Ammar Maraqa, Cisco's Chief Strategy Officer, put the security gap in operational terms: 60 percent of last year's breaches exploited vulnerabilities in systems that had not been patched, either because patching required downtime the organization could not accept, or because the infrastructure was too old to receive the patch at all. AI-powered attack tools now compress the window between vulnerability disclosure and active exploit from weeks to minutes. A patch cycle built for weeks is not a security posture for that environment.

Live Protect addresses this directly by applying compensating controls at runtime, with no reboot and no upgrade required. The device keeps running while the protection activates. For infrastructure that cannot tolerate downtime, that changes what is operationally possible between disclosure and patch.

The deeper security architecture Cisco is building around agents has three components. The pending acquisition of Asterix extends zero-trust principles from human accounts to every agent, service, and process in the environment, covering what the industry calls non-human identity. AI Defense, expanded at Cisco Live, now covers agent supply chains and third-party agent platforms with inline behavioral visibility. And the Galileo acquisition, integrated into Splunk Observability Cloud, adds agent evaluation and behavioral assessment on top of infrastructure performance monitoring.

The phrase that captured the architecture across multiple sessions was the shift from access control to action control. Traditional zero trust asks whether a credential is valid. Action control asks whether what the agent is doing right now, with this specific tool, for this specific task, is within the scope that was intended when it was deployed. Patel was direct that rules-based systems cannot hold that boundary, because agents find their way around static rule sets. The requirement is a dynamic model that intercepts anomalous behavior in real time.

The G300 chip, announced at Cisco Live, delivers 102.4 terabits per second of ethernet switching at a three-nanometer process node, with 246 billion transistors in a 100-square-millimeter package. The P200 handles scale-across networking, connecting data centers hundreds of kilometers apart so they operate as a single logical compute unit. Cisco also announced that Silicon One is now integrated into the Nvidia Spectrum X architecture, as the only non-Nvidia silicon participating in that scale-out cluster design, with the Nexus 9100 now shipping and Spectrum Six development underway.

These are not separate hardware announcements. They are the physical layer on which every Cloud Control and AgenticOps capability runs. Software built on faster, purpose-designed silicon performs differently than software running on general-purpose infrastructure, and the gap compounds as agentic workloads scale.

Cisco IQ, which reached 1,700 customers onboarded within weeks of launch, now includes Quantum Ready Assessments. Anthony Grieco, Chief Security and Trust Officer, described the threat as harvest-now-decrypt-later: encrypted data being collected today that becomes readable once quantum computing reaches sufficient capability. Most organizations do not know they are already targets of this approach. The assessment identifies which cryptographic assets are most exposed and where to begin modernization. Cisco has committed to quantum-safe communications across most core products by December 2026.

Token cost ran as a parallel theme through several sessions. Patel raised it directly: a single agent operating without governance on an unconstrained authorization scope can exhaust a material portion of an organization's AI budget in days. Cloud Control's observability module, powered by Galileo and Splunk, shows token consumption by agent in real time and allows an operator to terminate a runaway agent before the budget impact compounds. Patel described the industry as still in phase one, where organizations are learning rather than scaling. Governance built now costs less than the conversation with a CFO after the first surprise.

About Shashi Bellamkonda

Loading bio…

LinkedIn Profile

Get new posts by email:

Subscribe

Powered by