For the last decade and a half, the way you ran your network looked more or less the same. You logged into a console. You clicked through a dashboard. You made a change, checked a status, pulled a report. The GUI was the front door to almost everything.
That front door is about to close.
In its recent research, The Future of NetOps Is Agentic (January 2026), Gartner® projects a significant shift in how network and network security work gets done. Today, roughly 70% of administrative activity across firewalls, SSE, SASE, and switching still originates in a vendor’s dashboard. In this report, we see that by 2030, Gartner expects that number to fall below 30%, with AI agents becoming the single most common way network activities get executed. These are true agents that act on their own, a category well beyond the chatbots and assistants most teams know today.
That’s a big claim, and an exciting one. If you’ve spent any time operating a real network, though, your first instinct probably has less to do with the promise of less clicking and more to do with a harder question: what happens when something goes wrong at machine speed, and no one was watching the dashboard?
That instinct is the right one. The move to agentic NetOps is coming, and the benefits are real. The hard part will be governing it, and that work deserves attention now.
Agents are not assistants, and that distinction matters
It’s worth being precise about what’s actually changing, because the industry has been loose with the language.
A chatbot answers a question, such as how do I add a VLAN? An assistant follows an instruction, such as provisioning the ports and adding the VLAN. An agent pursues a goal. It notices a latency problem, investigates the likely cause across multiple systems, proposes a fix, and, if you let it, implements that fix without anyone touching a console.
Gartner draws exactly this line. Chatbots are response-focused. Assistants are task-focused. Agents are goal-focused, interacting with their environment and adapting to achieve an outcome. The practical difference is autonomy. An agent can operate without a predefined workflow and without a human in the loop for routine work.
That autonomy is precisely what makes agents so valuable. A human can reasonably check logs on 15 to 30 devices in an hour. An agent can check over a thousand in a few minutes, cross-reference them against baselines and change history, reason about root cause, and recommend an action. It can do all of this around the clock, without getting tired or distracted. For teams stretched thin by a skills shortage, that kind of capacity becomes a genuine lifeline.
The same properties that make agents powerful also make them risky when they run ungoverned.
The new risk surface
When you shift the source of network change from a human clicking a button to an agent pursuing a goal, the risk doesn’t disappear. It moves. Gartner is candid about this, and the risks it names should sound familiar to anyone who has watched policy drift accumulate over the years.
Agents can take the wrong action confidently. Most are underpinned by generative AI, which means they can hallucinate, reason incorrectly, or miss context, and then act on it. Techniques exist to reduce hallucination, though it can’t be mathematically eliminated. A wrong change at machine speed can become an outage before anyone notices.
Agents can be inconsistent. Give the same agent the same input twice, and you may not get the same behavior. In a network, inconsistent behavior is another name for configuration and policy drift, the slow, quiet misalignment that compromises the stability of an environment and that most teams already struggle to catch.
Permissions get complicated fast. An agent’s access can’t simply be modeled on a human user’s access. There’s real complexity in deciding what an autonomous agent should be allowed to touch, at what level of granularity, and under what conditions it needs to stop and ask.
The trust boundary moves, too. Autonomous agents introduce new trust, risk, and security management challenges that conventional controls weren’t designed for. In a multi-agent environment, those risks compound. Gartner is clear that “Within most enterprise networks, there will be multiple agents (multiple agents per vendor and multiple vendors). Thus, there will not be one agent but a fleet of agents with specific roles/tasks.”
None of this argues against agentic NetOps. It argues for governing it deliberately from the start.
From “in the loop” to “on the loop”
Gartner frames the human shift neatly. Network operators move from being in the loop on every change, to being on the loop, where they supervise, audit, and set guardrails, and ultimately, for well-understood tasks, out of the loop entirely.
That’s a meaningful change in the job. The operator becomes an orchestrator. You’re no longer making every change by hand. You’re defining the boundaries within which agents are allowed to act, and you’re accountable for auditing what they did and why.
For that model to work, a few things have to be true. You need to be able to see what an agent changed, in context, across every enforcement layer. You need the agent’s reasoning to be explainable and traceable, so a proposed root cause or action comes with the “why” attached. You need granular guardrails, meaning the ability to decide, per device or per policy, whether an agent can make a production change on its own or has to check first. And you need a governed change process that treats an agent’s actions with the same rigor as a human’s, including risk analysis, documented approval, and a clear audit trail.
Notice that none of these requirements are new. They’re the same disciplines that separate a well-run network from a fragile one today: visibility, context, governed change, continuous validation. Agents raise the stakes on getting those disciplines right.
What this means for security teams now
The Gartner guidance to infrastructure and operations leaders is practical, and it’s a reasonable roadmap regardless of which vendors you use.
Use traditional automation for what it’s good at: the structured, predictable, deterministic workflows where a known script does the job reliably every time. Reserve agents for the unknown and the ambiguous, where their ability to reason across messy inputs actually earns its keep. Pilot agent technologies before you trust them in production. Validate their root cause analysis and recommended actions, and talk to reference customers. And before any agent goes live, insist that it can be governed, with extensive logging, explainability, granular guardrails, and integration with the ticketing and change systems you already run.
Underneath all of it sits a quieter prerequisite that’s easy to overlook. Your data has to be trustworthy. An agent reasoning over an incomplete or inconsistent picture of your network will make confident decisions on a shaky foundation. Good outcomes come from good data, meaning a normalized, accurate understanding of what’s connected to what, what’s reachable, and whether that matches your intent.
That’s the throughline that connects the agentic future to the work security teams are already doing. The enterprises that adopt agents safely will be the ones that already have visibility across a hybrid, multi-vendor environment, a governed process for change, and continuous validation that posture still matches intent. Agents make that foundation more valuable.
At Tufin, this is the problem we’ve spent years working on. We give security teams a unified, trusted view of connectivity and a governed way to manage change across firewalls, cloud controls, and segmentation. The agentic era only strengthens the case for that foundation.
The dashboard may be on its way out. The need to see, govern, and prove what’s happening on your network is only getting stronger.
Want the full picture on where NetOps is heading? Download the Gartner report, The Future of NetOps Is Agentic, to see the complete research and what it means for your team.
Source: Gartner, The Future of NetOps Is Agentic, Andrew Lerner, Mike Leibovitz, et al., 5 January 2026.
GARTNER is a trademark of Gartner, Inc. and/or its affiliates.
Ready to Learn More
Get a Demo
