The Network Is Ready for Agents. Are Network Operations?

For most of the history of network operations, software has waited for us. We investigate an alarm, open a ticket, decide what needs to happen, and then a person, script or automation executes the change.

We have made this process significantly faster with APIs, automation and orchestration. But the fundamental relationship has remained largely the same: humans decide, software executes.

Agentic NetOps has the potential to change that.

Gartner recently described Agentic NetOps software as a disruptive, early-stage market that is redefining how network operations decisions are planned, governed and executed. I think they are right about the scale of the change. But the most interesting part of Agentic NetOps is not the agent itself. It is everything the agent will need around it.

From executing commands to pursuing outcomes

Traditional network automation is deterministic. Create this VLAN. Configure these ports. Allocate this address. Push this configuration. Somebody still has to decide which automation should run.

Agents introduce a different abstraction. Instead of describing every step, we can increasingly describe an outcome: “Prepare this site for 150 employees.” “Find out why connectivity at this location has degraded.” “Replace these switches without violating our change policy.”

The interesting part is not that we can type these requests into a chat interface. It is what could happen next.

An agent could determine what information it needs, retrieve it from multiple systems, reason about dependencies, select the appropriate workflows, request approval where necessary, execute permitted actions and verify the result.

That is more than task automation. We are starting to automate parts of operational decision making.

Intelligence is only part of the challenge

Imagine asking an agent a relatively simple question: “Can we replace this switch tonight?”

To answer properly, it needs to know far more than the configuration of the switch. Which site is it in? What depends on it? Which networks does it support? Is there an approved change? Is tonight inside the permitted change window? What does a successful replacement look like?

A network engineer can answer these questions because they understand the environment around the device. An agent needs that environment represented in software.

This is why I believe operational context will become one of the most important foundations of Agentic NetOps.

Networks are rarely operated from one system. Monitoring knows whether a device is healthy. IPAM knows which addresses are allocated. ITSM knows which changes have been approved. Controllers know what is configured. Documentation and CMDB systems contain another part of the picture.

An intelligent agent sitting on top of another silo does not solve this problem. It simply gives the silo a better interface.

At NetSymphony, we have spent a lot of time on this challenge. We connect infrastructure and operational systems around a common model of the network, bringing together devices, locations, networks, organisational structures, IP information, monitoring data and workflows.

Today, that context helps people operate complex environments. Tomorrow, I believe it will also help agents understand them.

Autonomy needs guardrails

Giving software the ability to reason about a network is exciting. Giving it the ability to change a production network should make us considerably more cautious.

An agent needs to understand not only what it can do technically, but what it is allowed to do operationally. That means permissions, policies, approvals, validation and auditability.

Gartner also points toward guardrails and governance as important characteristics of this emerging category. At NetSymphony, we already approach automation with this principle in mind. Change Guard provides a controlled layer around network changes, helping ensure that automation operates within defined policies and boundaries.

That principle becomes even more important with agents.

I don't think the future is an AI agent with unrestricted access to every router, switch, firewall and controller. A more realistic future is graduated autonomy. Low-risk, repeatable operations might happen automatically. Others could be investigated and prepared by an agent but require human approval before execution.

Trust will be earned operation by operation.

Agents should not inherit our silos

We have spent decades dividing IT into domains: network operations, security operations, service management, infrastructure and cloud. But operational problems rarely respect those boundaries.

A device behaving unexpectedly might involve networking, configuration and security. A new site might require IP allocation, network configuration, security policies, equipment orders, monitoring and documentation.

Humans experience these as one operational problem. Our tools often experience them as six separate ones.

Gartner has already pointed to Agentic NetOps as a potential enabler for closer NOC and SOC convergence. I think the implication is broader. With the right context and approved capabilities, agents could reason across operational domains rather than forcing people to translate between them.

That may ultimately be more valuable than automating any individual task.

The orchestration layer becomes more important, not less

There is an assumption that sufficiently capable AI will make traditional infrastructure software less relevant. I suspect the opposite will happen in network operations.

An AI agent should not need unrestricted access to ten infrastructure systems and then be trusted to figure everything out. It needs an environment through which it can understand the network, discover approved capabilities, respect permissions and trigger controlled workflows.

This is where I see an interesting evolution for platforms like NetSymphony. We have been building an operational layer that sits across vendors and existing systems, maintains context and turns network processes into governed workflows.

We originally built that architecture for people. Increasingly, I think it is also the kind of architecture agents will need.

The model provides the intelligence. The operational layer provides the context and the ability to act. Governance determines how far it is allowed to go.

The real opportunity

I don't think the end state of Agentic NetOps is a network operations centre where AI silently runs everything and network engineers have disappeared.

The more compelling future is one where people are no longer the integration layer between systems. Engineers spend less time looking up information, coordinating repetitive tasks and moving changes through different tools, and more time defining how the network should operate.

We spent the last decade teaching software how to execute network tasks. Now we are beginning to teach software how to reason about why those tasks need to happen.

The agents will become more capable. The bigger challenge, and perhaps the bigger opportunity, is building the operational environment we can actually trust them to operate in.