Rethinking Network Infrastructure for Agentic AI Needs

Organizations are investing heavily in GPU technology, servers, and computational resources to advance AI capabilities. However, as agentic AI becomes more prevalent, it presents unique challenges for network infrastructure.

Unlike traditional chatbots only responding to inquiries, agentic AI systems can take on tasks autonomously, engage with various applications, and make decisions without human intervention. This shift means that much of the work will occur independently of a user at a keyboard.

According to the findings from Google Cloud’s State of Infrastructure in the Agentic AI Era, a significant 83% of companies recognize the need for network upgrades to effectively utilize production-grade autonomous systems. While computational power is critical, enhancing the network is equally vital as it must accommodate workflows that differ from standard human activity.

The New Era of Networking

For years, enterprise networks have been structured around conventional working patterns, where employees connect to applications and access data primarily during typical business hours. Traffic patterns were largely predictable, influencing how network architecture was built and managed.

However, the rise of AI and agentic systems disrupts this norm. Agents execute tasks, utilize APIs, engage data sources, and interact with other agents at any time, creating continuous, dynamic network traffic. As the reliance on these systems grows, networks must adapt to handle increasing complexity alongside parameters like performance, latency, resilience, and security.

These evolving demands reveal the limitations of traditional network operations (NetOps). Established models depend on monitoring tools to detect issues, with human operators stepping in for diagnosis and resolution. Future NetOps must transition to a comprehensive understanding of application and agent interdependencies. This requires greater visibility, AI-driven insights for correlation and root cause analysis, and automation to proactively address and optimize network issues.

The vision is to evolve from networks supplemented by AI help to those that operate with AI-driven intelligence and autonomy. This shift enables network teams to focus less on troubleshooting frequent issues and more on defining policies and refining architectures. The change isn't merely about keeping networks active around the clock; it's about redefining their design, security measures, and operational frameworks to sustain a continuously active environment of intelligent systems.

Scaling from Individual Tools to Networked Agents

Currently, many organizations manage AI through simple tools like copilots, assistants, and AI applications scattered across different business sectors.

However, research from IBM indicates that only 11% of leaders feel fully equipped for the anticipated scale of AI agent deployment in the upcoming year. As these deployments scale, network teams will confront an influx of activity, necessitating efficient management without a proportional increase in workforce.

To streamline operations, it is essential to minimize the time engineers waste on seeking information through dashboards, collecting data, and escalating routine inquiries. Providing operational insights and troubleshooting resources at critical moments can facilitate knowledge sharing among teams, avoiding overreliance on a few specialists. The expertise of personnel can then be directed towards making informed decisions about policy creation, understanding limits, and assessing risks.

Integrating Networking into AI Strategies

While AI development often prioritizes model design, application functionality, data management, and computational resources, it is crucial that networking is incorporated into these discussions from the outset. Agents require context to deliver valuable insights, which means network operations should integrate information about users, devices, and applications rather than focusing solely on isolated alerts or metrics.

This comprehensive understanding allows teams to swiftly pinpoint issues and respond effectively. Human oversight continues to play a critical role, especially as AI agents gain autonomy. Implementing well-defined boundaries on agent capabilities and establishing controls to enforce these limitations is essential. Furthermore, teams must understand the rationale behind agents' recommendations to make informed decisions regarding their actions.

According to Gartner's predictions, by 2029, over 70% of organizations deploying AI agents in IT operations will encounter significant challenges related to service quality, security, or costs due to inadequate controls. This highlights the necessity of embedding visibility, automation, reliability, and control into AI infrastructure frameworks from the beginning.

Establishing Infrastructure for a Constantly Active AI Enterprise

The emergence of agentic AI is reshaping the operational use of enterprise networks. Activity is becoming less dependent upon employee presence online, and network teams must manage these changes while avoiding an increase in the operational workload.

Enterprises need not completely revamp their networks; however, they should assess existing vulnerabilities in their infrastructure as AI adoption grows. Particular focus should be on areas lacking visibility for quick problem identification, where response times lag, or where routine maintenance still demands considerable manual effort.

A network designed around traditional 9-to-5 hours must evolve to enable seamless collaboration between humans, applications, and agents who operate at different tempos and schedules. Successfully realigning this infrastructure will yield tangible benefits in areas such as reducing disruptions, accelerating resolution processes, enhancing user experiences, and minimizing time spent diagnosing issues.