Cisco is betting that the next job for AI agents inside the enterprise won't be drafting marketing copy or summarizing reports — it will be troubleshooting the network itself.
Speaking at Cisco Connect Taipei 2026 on July 7, Cisco Taiwan Chief Technology Officer Ivan Hsieh told attendees that agentic AI has moved past the hype stage. "It's not coming — companies are already living inside it," Hsieh said, arguing that IT operations are shifting from a model where humans hunt for problems to one where AI consolidates data and delivers a diagnosis.
Agent Traffic Is Straining Networks Built for Humans
Hsieh pointed to Cisco research showing that a single AI agent completing a task can generate roughly 450% more network traffic than a human doing the same job. That surge, he said, puts real pressure on enterprise infrastructure: when bandwidth is too narrow or link quality too poor, the result is latency — and that lag becomes a bottleneck for further AI adoption rather than an enabler of it.
Cisco's answer to that pressure, unveiled among a batch of AI announcements at Cisco Live, is Cisco Cloud Control, a management platform built on Cisco Data Fabric. According to the company, the platform gives operations teams a single management interface and a unified data layer, pulling together Cisco's networking, security, compute, visibility and collaboration products into one environment. The pitch is that AI agents and human engineers end up working from the same operational context, the same signals and the same action framework — rather than each department staring at its own dashboard.
Hsieh said Cisco Cloud Control isn't limited to Cisco's own stack. It's designed to plug into third-party AI ecosystems, letting enterprise customers bring their own agents or build custom workflows on top of it. For the engineer on the ground, he said, the experience resembles using any generative AI chat tool: a single conversational window where staff can ask questions in plain language, while the system pulls data from multiple platforms behind the scenes to help pinpoint what's actually wrong.
A Slow Conference App Becomes the Test Case
To illustrate how that plays out, Hsieh used a familiar office headache: a screen-sharing app in a meeting that suddenly runs slow. Historically, he said, that kind of complaint sends network, security, application and conferencing teams off separately to check their own systems — and it often ends with every team concluding, "not my problem," while the root cause stays hidden.
With Cisco Cloud Control, Hsieh said, an engineer can put that same question directly to an AI agent. The system cross-references network, meeting-platform, security and application data at once. In the scenario Hsieh described, the agent might determine the lag isn't a network issue at all, but a security-scanning process that stalls midway through a file-sharing step, degrading the user's experience.
The shift, Hsieh argued, is not that AI replaces IT staff, but that it compresses the time spent chasing a problem across departments. Engineers still confirm the diagnosis and decide how to respond — but they start from a synthesized picture instead of a blank investigation.
Why General-Purpose Chatbots Aren't Built for Network Duty
Hsieh also made a case for specialization over generality. Widely used models such as ChatGPT, Gemini and Claude are broad, general-purpose tools, he said, but enterprise IT operations demand something narrower: models with lower latency and a working understanding of networking and security context specifically.
Drawing on more than four decades of networking data, Hsieh said Cisco has trained what it calls deep network models, along with task-specific models tuned to particular product lines — including dedicated models for security analysis and for meeting-experience quality. When a customer submits a request, the system breaks it down and routes pieces of the task to the relevant specialist model, aiming for both speed and accuracy in that specific operational context.
That layered approach is central to what Cisco calls AgenticOps — a framework the company positions as more than a chatbot bolted onto an IT help desk. The goal, according to Cisco, is to turn network, security, visibility, application and collaboration data into signals that AI can actually interpret and act on, moving operations teams from reactive alert-handling toward proactively surfacing problems before they escalate.
Splunk Supplies the Data Plumbing Behind the AI Layer
None of that works, Cisco argues, without the underlying data being properly governed and unified — which is where Splunk comes in. Cisco Vice President Yueh-Tien Lin noted that Cisco's acquisition of Splunk more than two years ago added data-platform, SOC and SIEM capabilities that, combined with earlier observability acquisitions such as AppDynamics, were meant to give enterprises end-to-end visibility across their environments.
Cisco says its Data Fabric, built on the Splunk platform, spans networking, security, applications and infrastructure, converting large volumes of scattered data into intelligence that AI systems can act on — intended to help security teams run operations more efficiently.
The larger implication, Cisco suggests, is that adopting AI agents isn't primarily a question of which model a company uses. It's whether that company's data can be integrated, put in context and turned into decisions an agent can actually act on. For IT teams, that reframes the job ahead: less time spent manually checking systems one by one, and more time spent supervising AI agents, verifying their recommendations, issuing response strategies and confirming that whatever the AI flagged actually got fixed.
