"Network traffic generated by artificial intelligence (AI) agents can increase by up to nearly 400 times compared with before. The resulting infrastructure bottlenecks will become increasingly apparent."
DJ Sampath, senior vice president of AI software and platforms at Cisco's security business, stated accordingly at the "Cisco Live US 2026" media briefing held recently. As AI evolves from simple Q&A chatbots into agents that directly carry out tasks, he said corporations need to redesign their network, security, and data center infrastructure.
Cisco assessed that the spread of AI agents not only changes how applications are used but also demands an overhaul of networks, security, data centers, and collaboration systems across the board.
◇ "A major shift from chatbots to agents… infrastructure bottlenecks will grow"
Sampath said, "A sweeping transition is happening now from chatbots to agents," and "Agents are not just tools but are emerging as colleagues that perform work together." He cited infrastructure constraints, lack of trust, and data gaps as key challenges in the AI agent era.
He explained that as AI agents increase, the complexity of corporations' infrastructure will also grow. Sampath said, "Routing which task to which model, which action, and which tool is extremely complex," and "Every action an agent takes becomes a telemetry event, so governance is needed to manage and control this."
Cisco's proposed solution is "Cisco Cloud Control." It manages network, security, data center, and application operations data on a single platform, with AI agents suggesting root-cause analyses and response options. Instead of jumping between multiple dashboards, operators review the agent's analysis and recommendations and approve necessary actions.
Sampath said, "Cisco Cloud Control is a unified, AI-native management platform that allows people and agents to work together," adding, "Through AI Canvas, users can obtain needed information by conversing with security products, firewalls, and networking gear."
Cisco also introduced "Cloud Control Studio," which lets corporations build custom agents and apps. It includes an "Agent Builder" for customers to develop and train their own agents and an "App Builder" to create applications tailored to workflows. Sampath said, "Customers say customization to their own environments and workloads is crucial," and "Agent Builder trains agents to carry out specific tasks of the customer organization."
OpenAI Codex is also integrated into App Builder. Sampath said, "We will embed OpenAI Codex within the Cisco Cloud Control platform so customers can build custom apps and workflows suited to each environment," adding, "From the initial design stage, you can implement applications and automation with observability and security."
◇ "Decades-old networks cannot support corporations' AI ambitions"
Ben Dawson, president for Asia-Pacific, Japan, and China (APJC) at Cisco, emphasized modernizing infrastructure for the AI era. Dawson said, "The AI revolution has already arrived, and digital infrastructure has become a foundational driver of economic and business outcomes," adding, "Networks built decades ago cannot support corporations' AI ambitions."
Dawson said, "Customers keep talking about the need for network modernization," and "Infrastructure must be built with security in mind from the start." He said what's needed is a full-stack infrastructure where networking, computing, storage, security, and observability work together, rather than tacking security on later.
Cisco also assessed that the focus of AI investment discussions is shifting. Dawson said, "In the past, the main conversation was how to build AI capabilities; now the question is how to achieve tangible return on investment (ROI)," adding, "The conversation is moving from CIO-centered to CFO-centered."
He said, "In the AI era, networks, security, data, observability, and collaboration must function as one system," adding, "Cisco will support customers' transitions as a core infrastructure company for the AI era."
◇ "AI also speeds up attacks… the way we patch vulnerabilities must change"
Tom Gillis, senior vice president and general manager of Cisco's infrastructure and security group, stressed that AI is also increasing the speed of security threats. Gillis said, "As frontier models advance rapidly, the way we operate and protect infrastructure at scale is fundamentally changing," adding, "The old approach of finding a vulnerability, patching it, and then leaving it for a year, 18 months, or two years no longer works."
He said, "As automation advances, attackers are not only finding vulnerabilities but also quickly acquiring the means to exploit them," and "The gap between the time a vulnerability is found and the time it is actually exploited is narrowing." Accordingly, he said, security features that precisely control specific processes or file access in runtime environments are needed.
Access control for AI agents was also cited as a new challenge. Gillis said, "Access control for people and access control for agents must be distinguished," adding, "We need to identify whether the entity typing on the keyboard is a person or an agent acting on a person's behalf."
He likened agents to a company "intern." To perform tasks, they need email, internal tools, and data access rights, but excessive privileges increase security risk. Gillis said, "Agents should be given a nonhuman identity separate from people, and their access should be precisely controlled in terms of which resources they can access, for what tasks, and for how long," adding, "You must manage not just simple access control but also what agents are allowed to do."
◇ "Operators will take on the role of supervising AI agents"
Anurag Dhingra, senior vice president and general manager of Cisco's enterprise connectivity and collaboration division, pointed to secure connectivity and collaboration tools as requirements for workplaces in the AI era. He said, "A future-ready workplace needs both secured connectivity and effective collaboration tools," adding, "Video conferencing and collaboration tools like Webex and network management capabilities must evolve together."
Dhingra also expects AI agents to play a larger role in network operations. He explained, "In Cisco Cloud Control, operators will take on the role of supervising the work agents perform," and "When an agent proposes an optimization, operators can review the network brief, root-cause analysis, and risk score to decide on actions."
Digital Twin was also mentioned as a key tool for AI network operations. Dhingra said, "You can identically implement networking gear such as routers and access points in a virtual environment to test," adding, "Before applying an agent's proposed action to the actual network, you can validate its impact in a digital environment."
A network strategy to connect multicloud environments was also presented. Dhingra said, "Corporations' applications are being built across multiple environments such as AWS, Google Cloud, Microsoft Azure, data lakes, and Active Directory," adding, "Consolidating and managing all these clouds is not easy."
To support this, Cisco plans to offer a "Multicloud Fabric" (a common fabric that consolidates, manages, and protects dispersed infrastructure as if it were a single network) that can run directly from Cisco Cloud Control. As AI agents move to the forefront of corporate work and infrastructure operations, networks are being redefined not as mere connectivity but as foundational infrastructure for the AI era that spans security, data, collaboration, and operations automation.