Networks used to have one job: stay up, says Sujai Hajela, HPE's executive vice president and general manager of campus and branch networking. But in today’s AI era, that's not enough, and manually managed networks simply can’t keep up with the growing scale, complexity and expectations around modern distributed network infrastructure.

Now, Hajela says, “up is not the same as good.” What organizations need are self-driving networks designed to track what users — from C-suite leaders in their offices to robots on factory floors — are actually experiencing, and use that insight to spot problems before they cause disruptions, adapt in real time and continuously optimize themselves.

As agentic AI technology becomes more mature, networks now have the ability to act automatically across campus, branch, routing and data center environments — at a scale human operators can’t match — ensuring both uptime and user experience keep pace, Hajela says.

“Up is not the same as good.” Sujai Hajela, Executive Vice President and General Manager, Campus and Branch, Networking, HPE

But building one takes more than deploying another AI tool. A truly self-driving network requires these five defining traits.

1. AI Efficacy Builds Trust

AI efficacy — consistently diagnosing problems correctly and knowing what action to take — is the foundation of trust in a self-driving network. “If you do not have proven efficacy, how do you trust the solution?” Hajela asks.

Hajela compares it to a self-driving car. You may understand how the technology works, but when your car needs to turn left in front of oncoming traffic, you need confidence it will not only perform the action but cross the road without incident. “You cannot afford false positives,” he says. “Imagine you let the network self-drive without efficacy. It can bring your network down.”

AI efficacy takes time to build. HPE has spent more than a decade turning real-world troubleshooting experiences into reusable intelligence for its AI engine, Marvis AI. “Today, Marvis is able to answer over 80% of the wireless problems reported by the user,” he says.

2. AI-Native Architecture Provides The Right Foundation

Too many vendors simply layer large language models onto legacy infrastructure, but Hajela argues that’s not enough. An AI-native network is built from the ground up to collect and understand user-experience data, learn from past problems and fixes, and give AI the framework it needs to act.

That means curated live data that quantifies user experience from many sources (from network devices and applications to clients and devices), a memory of past problems and their resolutions and synthetic data to predict future problems. In HPE’s AI-native architecture, that information is collected, analyzed and stored in a microservices cloud to deliver scale, agility and resiliency, Hajela says. “You cannot deliver self-driving functions, such as healing, optimization and protection, without the right foundation.”

From there, the system can recognize a problem, draw on what it has learned and take corrective action.

3. Autonomous Self-Driving Operations Enable IT To Deliver More Value

A truly self-driving network doesn’t wait for a help-desk ticket. It automatically recognizes a problem, determines the appropriate response and takes action on its own, Hajela says.

The agent knows what needs to be done, but knowing how to do it requires capturing the experience and expertise of human operators and translating them into digital skills the agent can use. “We are digitizing our experience and making the agents think like humans,” he says. “That’s the critical part.”

“We are digitizing our experience and making the agents think like humans.” Sujai Hajela, Executive Vice President and General Manager, Campus and Branch, Networking, HPE

As more network operations become autonomous, IT teams can step back from constant troubleshooting and focus on shaping how the network serves the business. “The role of IT is changing because they are now truly becoming a business asset rather than a call center or an IT asset,” Hajela says. “They are translating the business needs into how the network should work for them.”

4. Integrated Security Means Users, Devices And Workloads Are Always Protected

Integrated security means embedding zero trust — never automatically trusting a user or device — into the network itself with a common policy protecting users and devices across both north-south and east-west traffic, Hajela says.

Traditional security models have long focused on the perimeter, using firewalls and cloud-based services to protect north-south traffic flowing in and out of the network. But as AI and agents reshape network traffic, east-west communication between devices within campuses and branches is growing, creating challenges that legacy perimeter defenses can’t address alone.

“Zero trust cannot be just about the security devices doing their job,” Hajela says. “The network has to come together, and that’s what we mean by integrated security.” That means enforcing a common security policy wherever users and devices connect, with automated protection to find and fix problems quickly. AI adds another layer, helping detect threats and respond faster.

5. Earned Autonomy Lets Organizations Move At Their Own Pace

Chief information officers and chief information security officers want the benefits of AI but with guardrails and a clearly defined scope, Hajela acknowledges. To get there, he recommends taking a gradual path to self-driving operations where the system earns trust before it earns the right to act on its own. “You don’t have to go to full autonomy overnight,” he says.

The system can first identify a problem, surface the evidence and recommend a solution for an operator to validate. Once the operator is confident the system can handle that specific issue, an action can become “trusted” with specific AI agents given permission to autonomously find and fix that specific issue — without humans in the loop

“Autonomy has to be earned,” he says. Over time, Hajela envisions humans moving from “in the loop,” approving individual actions, to “on the loop,” overseeing autonomous operations and ensuring the right outcomes were achieved.

Self-Driving At Your Own Speed

Looking three to five years out, Hajela believes self-driving networks and agentic AI are inevitable. “Whether you like it or not, self-driving networks are coming,” he says.

Done right, the rewards are substantial, Hajela says, including a 90% reduction in user-generated trouble tickets; faster network deployment, down from months to less than a day; and better business outcomes.

Getting there, however, takes time. Hajela’s advice to CIOs and CISOs is to start small and let the technology earn trust before expanding its role. And as AI takes on more routine work, he advises IT teams to prepare for a shift in responsibilities. Engineers may spend less time writing code and more time directing, validating and applying AI to solve problems.

Just don’t hand over control overnight, Hajela says. “Embrace the self-driving network at your own pace.”

Writer: Sarah Lindenfeld Hall