Network professionals are increasingly willing to hand over day-to-day network operations to artificial intelligence, according to a new survey of 1,000 IT and network operations leaders. The research, conducted by a major networking vendor and a technology research and advisory firm, found that 80% of respondents are comfortable granting AI a high or fully autonomous role in network operations. That comfort level comes with a major caveat: nearly all of them want strict guardrails in place before they let AI act on its own.

The findings arrive at a moment when enterprise networks have become too complex for many human teams to manage manually. The forces driving that complexity include cloud adoption, security integration, application performance demands, endpoint sprawl, and, notably, the rapid growth of AI itself. As organizations deploy more AI workloads, network traffic patterns are shifting, data volumes are rising, and the pace of change is accelerating.

Key Facts From the Survey

• 80% of network professionals are comfortable giving AI a high or fully autonomous role in network operations.
• 51% already use agentic AI tools in production to take corrective action in real time.
• 84% expect to reach a fully AI-led operating model within 12 months.
• 75% use AI in some fashion for network operations today.
• 56% will allow autonomous action only with human approval; 24% are comfortable with AI acting without human oversight.
• 82% are comfortable allowing AI to make some production network changes on its own for certain categories.
• 99% say they would not trust AI to act without guardrails such as explainable actions, human approval, policy-based limits, emergency overrides, role-based access control, and immutable audit trails.
• 59% change production network environments at least daily, and 57% say their change processes cannot keep up.
• 92% say performance issues cross multiple domains, including cloud, security, applications, and endpoints.
• 95% say existing non-agentic AIOps tools fall short, often because they require too much human interpretation and lack cross-domain visibility.
• The average organization generates about 4,100 monitoring alerts and events per day; 46% of alerts are closed without investigation.
• Mean time to resolve a network incident is 88 hours, while the median is 12.5 hours.
• Organizations rely on an average of 10 tools for end-to-end visibility, but those tools tend to be siloed.

Why Networks Have Become Too Complex for Humans Alone

The survey paints a picture of network operations under severe strain. More than half of respondents, 59%, report making changes to their production network environments at least daily. Half of those organizations make multiple changes per day, and for a meaningful share, change happens multiple times per hour. As a result, 57% say their change processes cannot keep up with the pace of business demands.

Complexity is also crossing traditional boundaries. Ninety-two percent of respondents say performance issues tend to span multiple domains, including cloud, security, applications, and endpoints. That means a problem that begins in one area can quickly affect others, making diagnosis and resolution far more difficult. Existing non-agentic AIOps tools are not keeping pace. Ninety-five percent of respondents say those tools fall short in significant ways, mostly by requiring too much human interpretation and by lacking cross-domain visibility.

The generative AI boom is adding fuel to the fire. Two-thirds of respondents say the rise of generative AI has increased network complexity. Traffic analysis of direct-to-AI communications suggests average daily AI traffic is on a trajectory to double every six months. That acceleration is likely driven by the growing complexity of AI tasks, which demand more data exchange than a simple query. In other words, AI is both a solution to network complexity and a cause of it.

Alert volume tells a similar story. The average organization generates around 4,100 monitoring alerts and events per day, with 51% being network-related. A typical practitioner can review, investigate, and resolve about 21 network alerts a day, according to estimates from the research firm. That means it would take a team of about 100 specialists to handle the daily volume. Since few organizations have that much staff, nearly half of alerts, 46%, are closed without investigation.

Alert fatigue is a meaningful source of employee dissatisfaction for 65% of respondents, while 67% say alert volumes prevent teams from doing other critical work. The problem is not unique to networking. Similar research around security alerts has found that well under 50% of alerts are investigated. The common thread is a growing gap between the number of issues generated by modern infrastructure and the human capacity to address them.

The Push Toward Agentic Operations

The survey describes a transition from AIOps to agent-powered operations, or AgenticOps. Traditional AIOps tools mostly analyze data and offer recommendations. Agentic AI goes further: it can take corrective action in real time. Already, 51% of respondents use agentic AI tools in production for that purpose, rather than simply taking advice from them. Eighty-four percent expect to reach a fully AI-led operating model within 12 months.

Three-quarters of the 1,000 IT and network operations leaders surveyed say they use AI in some fashion for network operations. Eighty percent are comfortable granting AI a high or fully autonomous role. More than half, 56%, will do so only with a human approving those actions. But 24% are comfortable with AI taking network actions with no human oversight. The vast majority, 82%, are comfortable allowing AI to make some production network changes on its own for some categories.

Industry analysts who follow networking trends say the move toward agentic operations is inevitable. The combination of increasing network complexity and a decreasing number of humans with the right skills to solve networking problems, including cross-domain issues, is creating pressure that manual processes cannot relieve. One analyst compared the trust curve for agentic AI to autonomous vehicles, which get into far fewer accidents than cars with human drivers. The same logic suggests AI agents will make mistakes, but fewer than people will, and they will free human teams to focus on higher-value work.

Security operations may provide a preview. Faced with too many issues to handle manually, security professionals are increasingly automating response, even when that means shutting down a resource. The thinking is that the potential losses from inaction are greater than the potential impact to the business. Network teams appear to be moving in the same direction, albeit with more caution and more requirements for oversight.

Guardrails Are Non-Negotiable

The survey was released amid calls to slow AI adoption, partly in response to incidents in which autonomous AI agents have run amok. Yet hundreds of network professionals are willing to cede at least some control of their networks to AI agents. They will do so, however, only if strict guardrails are in place. Nearly every respondent, 99%, said they would not trust AI to act without such guardrails.

The guardrails respondents identified include:

• Explainable AI actions
• Human approval for actions
• Policy-based operational limits
• Emergency override mechanisms
• Role-based access control
• Immutable audit trails

Guardrails build trust in AI by letting users see the reasoning that leads an agent to a conclusion. At the same time, while AI agents can draw conclusions from the intelligence they examine, any actions they take map to predefined workflows based on the network team's standard operating procedures for how to respond in different situations.

Another key constraint is that agentic agents do not talk to other agents. That helps a great deal because when agents are constrained to figure things out themselves, they are less dangerous. Agents are also built around specific skills or knowledge and a defined scope of responsibility. When that approach is applied on a product basis, with tight integration and controlled boundaries, it gives enterprise network operations teams the confidence to leverage agentic AI to help them manage increasing complexity.

Alert Overload, Tool Sprawl, and Slow Incident Resolution

The operational data in the survey helps explain why so many network teams are willing to experiment with autonomy. Organizations rely on an average of 10 tools to try to maintain end-to-end visibility, but those tools tend to be siloed. That makes it difficult to diagnose problems that cross domains, as many do. The result shows up in resolution times.

The mean time to resolve a network incident is 88 hours, while the median is 12.5 hours. The mean is skewed by some organizations that take a week or longer to resolve issues, reflecting the degree of complexity. That gap between mean and median suggests a long tail of especially difficult incidents that consume disproportionate time and resources. It also points to an opportunity: having AI help with analysis and automate the root cause process could reduce both the frequency and the duration of major incidents.

One practical example is when an organization tends to take the same corrective action every time a situation occurs. If a team has done it 14 times, it can automate the response, but it should still notify humans that the action was taken. That kind of targeted automation preserves human oversight while eliminating repetitive work. It also builds a record of what AI did and why, which is essential for trust and auditability.

Vendor Response: A Unified AI-Powered Control Plane

In response to these pressures, at least one major networking vendor is proposing a new cloud control platform intended to provide a unified view and management plane for networking, security, compute, observability, and collaboration solutions. The platform also applies agentic AI to diagnose and resolve issues, including those that cross domains. It is, once again, the proverbial single pane of glass, this time with an AI twist.

It may seem somewhat ironic that a vendor that sells networking gear that has become too complex to manage is now also selling the solution intended to address that complexity. Analysts acknowledge that network vendors have been complicit in creating the problem. But they also say it is good to see vendors simplify things now. The network is being used in many more ways than ever before. It is more complex for a number of reasons, including that networks support orders of magnitude more devices and connect to pretty much everything.

The road to fully AI-led operations will not be instantaneous. Even among the 84% who expect to reach a fully AI-led operating model within 12 months, most will likely proceed in stages. Human approval will remain part of the process for many organizations, especially for high-impact changes. The survey suggests that the question is not whether AI will play a larger role in network operations, but how quickly organizations can put the right controls around it. For network teams facing relentless change, rising alert volumes, and cross-domain complexity, agentic AI is increasingly seen not as a luxury but as a necessary partner.

Source:Network World News