Enterprise network operations have officially entered a new era of transformation, driven by an unsustainable surge in complexity that has left IT professionals eager to hand over the reins to autonomous technology. According to a comprehensive new report released by Cisco and research firm Omdia, traditional methods of managing enterprise infrastructure are falling short, prompting a massive shift toward agentic artificial intelligence capable of taking real-time corrective actions with minimal human intervention.
The report, titled “The Impact of Agentic AI on Network Operations,” surveyed 1,000 IT and network operations leaders globally to uncover how modern digital environments are impacting daily workloads. The findings paint a striking picture of corporate networks that have expanded far beyond the capacity of human teams to manually monitor and maintain.
More than half of the surveyed professionals—51%—reported that they are already utilizing agentic AI tools in production environments to execute real-time corrective actions, moving well past the traditional phase of merely receiving recommendations from AI assistants. Furthermore, an overwhelming 84% of respondents expect their organizations to transition to a fully AI-led operating model within the next twelve months.
Three-quarters of the surveyed IT leaders confirmed they leverage artificial intelligence in some capacity for network operations. Meanwhile, 80% expressed a high level of comfort granting AI a “high or fully autonomous role” in managing network systems. While 56% of these forward-looking professionals still prefer a human-in-the-loop approach where staff must approve actions before execution, a notable 24% stated they are completely comfortable allowing AI to manage network operations with zero human oversight. Overall, 82% reported comfort in letting AI execute production network changes independently across specific categories.
Moving Toward AgenticOps
Cisco and industry analysts are categorizing this evolution as a definitive leap from traditional AIOps—artificial intelligence for IT operations—to agent-powered operations, commonly referred to as AgenticOps. Industry veterans view this transition not as a radical gamble, but as a necessary and logical evolution to keep pace with modern enterprise demands.
Zeus Kerravala, founder and principal analyst at ZK Research, emphasizes that the industry has reached a point where manual intervention is no longer viable. Network professionals will inevitably embrace agentic operations out of necessity, he argues. While building absolute trust in these autonomous tools will take time, Kerravala draws a parallel to the adoption curve of autonomous vehicles. Studies have consistently shown that driverless cars are involved in significantly fewer accidents than vehicles operated by human drivers.
By that same token, Kerravala notes that agentic AI tools will undoubtedly make mistakes, but they will make far fewer errors than human operators, while simultaneously freeing internal staff to concentrate on higher-value business initiatives.
Jim Frey, principal analyst for networking at Omdia, who collaborated on the development of the Cisco survey, echoes this sentiment, viewing the widespread industry pivot to AgenticOps as an absolute inevitability. The primary catalysts, according to Frey, are the relentless compounding of network complexity combined with a shrinking talent pool of human specialists who possess the cross-domain skills required to troubleshoot intricate networking problems.
AgenticOps Must Include Effective Guardrails
The release of the Cisco survey coincides with broader industry debates regarding the pace of AI deployment. These discussions have been heavily influenced by notable events, such as the Hugging Face security incident involving autonomous AI agents running unchecked. Despite such cautionary tales, hundreds of network professionals surveyed remain eager to delegate significant network control to AI systems, provided that rigorous safety measures are firmly established.
The research highlights that strict guardrails are non-negotiable for enterprise adoption. Nearly every single respondent—99%—indicated they would refuse to trust autonomous AI actions without proper operational guardrails in place, a critical safeguard missing during incidents like the Hugging Face breach.
Joe Vaccaro, senior vice president and general manager of network platform and assurance at Cisco, explains that guardrails serve to build essential trust by providing visibility into the underlying reasoning an AI agent uses to reach a conclusion. Additionally, while these autonomous agents possess the capability to analyze intelligence and draw conclusions, any subsequent actions they take are strictly mapped to predefined workflows. These workflows mirror the network team’s established standard operating procedures for handling specific operational scenarios.
Frey points out another vital architectural constraint that reduces potential risk: current agentic tools generally do not communicate directly with other agents. Confining agents to isolated analysis and independent problem-solving significantly limits their potential danger. Furthermore, these agents are typically engineered around specific, highly focused skills, deep domain knowledge, and rigidly defined scopes of responsibility. When deployed correctly through established product ecosystems, this modular design allows organizations to maintain strict oversight.
Driving the Need: Complexity, Rapid Change, and AI Itself
The underlying drivers pushing organizations toward AgenticOps are deeply rooted in the day-to-day realities of modern IT environments. Network complexity is accelerating rapidly, with 59% of survey respondents reporting that they introduce changes to their production network environments on a daily basis. Among those organizations, half execute multiple changes every single day, while a meaningful percentage implement changes multiple times per hour. Consequently, 57% of IT leaders admit that their existing change management processes can no longer keep up with the pace of business.
Compounding this operational friction, 92% of survey participants noted that performance issues routinely span across multiple technical domains, seamlessly bridging cloud environments, security layers, applications, and endpoint devices. Furthermore, 95% stated that their legacy, non-agentic AIOps tools fall short in critical areas, primarily because they demand excessive manual interpretation and lack comprehensive cross-domain visibility.
The rapid rise of generative AI workloads has added yet another layer of strain. Two-thirds of respondents confirmed that the generative AI boom has directly contributed to increased network complexity. Cisco’s proprietary traffic analysis, which monitors direct-to-AI data flows, indicates that average daily AI traffic is on a trajectory to double every six months. This rapid acceleration is largely driven by the sophisticated nature of modern AI tasks, which demand vastly greater data exchanges than traditional database queries or web traffic.
The operational toll of this environment is stark. The average enterprise surveyed generates approximately 4,100 monitoring alerts and events every single day, with slightly over half directly related to networking. According to Omdia’s operational estimates, a typical IT practitioner can thoroughly review, investigate, and resolve roughly 21 network alerts per day. Managing the daily volume of alerts would theoretically require a dedicated team of approximately 100 specialists.
Because few enterprises possess workforces of that scale, nearly half of all generated alerts—46%—are simply closed without formal investigation. This relentless volume creates widespread alert fatigue, which serves as a meaningful source of employee dissatisfaction for 65% of respondents. Moreover, 67% report that the sheer volume of alerts actively prevents their technical teams from focusing on strategic, mission-critical projects.
ZK Research’s Kerravala notes that these alarming statistics align closely with his independent research into enterprise security alerts, where well under half of all generated warnings receive proper investigation.
To maintain end-to-end visibility, organizations currently rely on an average of ten distinct monitoring tools. However, these tools remain heavily siloed, making it exceptionally difficult for IT professionals to diagnose and resolve complex problems that cross multiple architectural domains. This fragmentation is directly reflected in resolution times. The mean time to resolve a network incident sits at 88 hours, while the median is 12.5 hours. Frey explains that the higher mean average is heavily skewed by organizations that require a week or longer to untangle deeply complex incidents, highlighting a clear opportunity for AI to step in, streamline data analysis, and automate root-cause identification.
In highly repetitive scenarios where engineering teams execute the exact same corrective action every time a specific failure occurs, automation becomes an obvious win. If a recovery procedure has been executed manually a dozen times without issue, allowing an automated agent to handle it—while logging the action for human review—drastically reduces operational friction.
Industry observers suggest that network professionals may soon adopt strategies similar to their cybersecurity counterparts. Faced with overwhelming alert volumes, security practitioners have increasingly embraced automated responses, occasionally choosing to isolate or shut down compromised resources automatically. The prevailing mindset in security is that the potential financial and operational losses of a breach outweigh the temporary business impact of a preventative shutdown. Network teams appear to be slowly warming up to a similar philosophy regarding automated remediation.
The Solution: Unified Control and AI Integration
In response to these mounting challenges, infrastructure vendors are rapidly rolling out unified management platforms designed to consolidate fragmented toolsets. Cisco, for instance, has introduced its Cloud Control platform, intended to provide a centralized view and unified management plane spanning networking, security, compute, observability, and collaboration solutions. By integrating agentic AI directly into the platform, Cloud Control aims to diagnose and resolve cross-domain issues autonomously, offering the industry yet another iteration of the traditional single pane of glass, this time supercharged with advanced artificial intelligence.
While some irony exists in the fact that legacy hardware vendors who contributed to infrastructure complexity are now selling the software solutions designed to resolve it, industry analysts view the shift pragmatically. Networks are utilized in vastly more ways today than ever before, supporting exponential numbers of connected devices and spanning hybrid ecosystems.
As the enterprise landscape continues to evolve under the weight of generative AI and expanding digital footprints, the transition toward agentic automation appears less like a futuristic experiment and more like an essential survival strategy for modern IT operations.