Most telecommunications service providers already know how to manage changes to their network.

They have established change windows, built approved playbooks, and used network automation over many years to make the process more reliable. The challenging part isn't implementing the change, it's deciding what to change, where, and why—while alarms, vendors, and network domains keep multiplying.

Agentic operations can help. This isn't a new way to implement change, it's simply putting AIOps (AI for IT operations) style intelligence inside the processes that operations teams already trust.

The result is lower operating cost, faster resolution, better use of infrastructure, and a more stable customer experience—decisions that used to rely on extensive discussion and review can now happen inside a governed, human-in-the-loop workflow.

The bottleneck is the decision, not the change

TM Forum’s Autonomous Networks framework defines 6 levels of autonomy. They describe the evolution from human-driven operations toward increasingly intelligent, closed-loop, and ultimately fully-autonomous network operations.

In this framework, Level 4 is described as intent-driven, predictive, closed-loop operations. Most service providers still sit in partial or conditional automation (Levels 2–3) which includes scripts and playbooks that run well once a human has made a decision about what to do. The gap is the decision cycle, and that is where AIOps and agentic AI matter most.

That gap is getting wider. 5G standalone, cloud-native network functions (CNFs), multivendor environments, edge infrastructure, and quickly shifting traffic patterns make it harder to see which signals matter, what will happen next, and which response is safe. Traditional, reactive operations often struggle to keep pace. The implementation path is often already automated. The intelligence in front of it is not.

Putting better intelligence into that cycle means service providers reach TM Forum Level 4 faster, in selected domains, without throwing away the automation already in production. Level 5 isn't a rip-and-replace event, it's the same closed-loop pattern, expanded and more tightly governed.

What stays the same. What gets smarter.

The change implementation layer is often already established and can stay. For many service providers, Red Hat Ansible Automation Platform is that layer: human-readable, repeatable, auditable workflows across network and IT, including Event-Driven Ansible when a known condition should trigger a known job.

What we're adding now is a new layer of intelligence. Red Hat AI brings models and agentic workflows into operations so the system can correlate signals, predict what happens next, and recommend—or, within policy and after a human has approved, select—the next action. Agentic AI can work through multiple steps: observe, compare against intent, and propose a response. It still executes through the same automation contracts, the same role-based access control (RBAC), and the same audit trail.

That's the shift. Better decisions. Faster decision cycles. The network change itself still looks like the process operations teams already know.

An AI model that flags a problem is useful. A system that identifies the issue, chooses an approved response, triggers the existing workflow, and verifies the outcome is much closer to Level 4 autonomous network operations. The playbook did not have to change. The decision did.

Service providers shouldn't have to choose between the automation that already works and the intelligence their network now needs.

A multi-vendor AIOps path, not a single-vendor brain

Autonomy that only works inside a single vendor domain is not autonomy service providers can use. Radio access networks (RANs), core, transport, IT, and cloud already come from different vendors. The decision will cross those domains. The change still has to land on each vendor's interface through the same disciplined automation path.

That's why the intelligence layer and the execution layer need to sit where vendors connect, not inside any one toolchain.

Red Hat AI includes an AI gateway function, which acts as a control plane managing and governing agent traffic. It manages policies, quotas, authentication, and governance—essentially deciding how and where models are served. As more models enter operations—fault correlation in the RAN, capacity prediction in the core, assistants in the network operations center (NOC)—service providers need one place to apply security policy, content controls, and usage governance. The gateway gives service providers a way to let models from various sources participate in the same decision process without a different trust model for every vendor domain.

The same logic applies after the decision. Security, observability, and policy cannot live only in a single vendor's tool. They need to operate at the platform layer, defining who is allowed to act, what they are allowed to change, and how every action is recorded. Ansible Automation Platform already does that for execution. Pair it with platform-level inference governance and service providers have a path to multivendor network automation—each domain can keep its specialist systems, and the service provider keeps one decision-and-control plane.

That is also the practical path to Level 5. Full autonomy won't arrive as a single-vendor closed architecture. It will arrive when intent, policy, and execution work across vendors, and when operations teams can still see and control what the system is allowed to do.

Fault management and self-healing networks

Fault management is one of the clearest places for service providers to start building AIOps and agentic AI capability.

Today's networks generate enormous volumes of alarms, events, logs, and telemetry. Operations teams can drown in signals causing classic alert fatigue. The challenge is multifaceted. The system needs to detect that something has gone wrong and also know which signals matter, what the root cause may be, what the impact could be, and what to do next.

AI-driven alarm correlation spots patterns, predicts failures, and supports root-cause analysis (RCA). That's the decision getting faster and better—and a direct path to reducing mean time to repair (MTTR).

Ansible Automation Platform runs the remediation service that many providers already trust: restart a service, isolate a path, open a ticket, verify the service. Same workflow. Same controls. Less time spent deciding under a flood of signals.

That is closed-loop fault management and self-healing network operations: detect, analyze, decide, remediate, verify. TM Forum treats fault management as a high-value Level 4 scenario. It serves as a good first domain for a practical reason: service providers aren't asking the NOC to change how change is implemented, just to put more intelligence in front of the playbooks it already runs.

Predictive capacity and maintenance

Capacity management is the same pattern in a different domain.

Traditional planning often relies on historical trends, periodic reviews, and conservative build-out. Demand doesn't wait. New services, events, application workloads, and geographic shifts can move it quickly.

AI can read historical and live telemetry, identify demand forming, and predict where capacity will break, helping provide predictive maintenance and predictive capacity planning before the customer feels it.

The implementation can stay familiar: provision, reconfigure, reallocate, or trigger the workflow that already exists. Predictive capacity is a faster, better planning decision connected to the network automation service providers already use to add or shift resources.

Level 4 in the domains that become problematic first

The journey toward autonomous intelligent networks won't happen overnight. Service providers have different starting points, architectures, vendors, and operating models.

That is why an open, modular approach matters. Service providers shouldn't stand up an isolated AI system or another operational silo. They should introduce agentic AI and AIOps into workflows they already run, and connect them to the automation investments already established.

The goal is to implement TM Forum Level 4 in the domains that cause problems first, and establish a foundation that can evolve to Level 5 later.

TM Forum Innovate Americas 2026 is built around that practical shift: moving agentic AI and AIOps into day-to-day network operations, and moving beyond pilots into production. The event takes place October 6–7 in Dallas, Texas. (TM Forum)

See it in practice at Innovate Americas 2026

Want to see how agentic AI and AIOps improve the decision process in a real network operations workflow?

Join Red Hat at TM Forum Innovate Americas 2026, and let's discuss how AI agents and agentic AI improve fault management, alarm correlation, and predictive capacity—while Ansible Automation Platform keeps executing the change the way service providers already do, across a multivendor network.

Read more about the session and add it to your Innovate Americas agenda: Red Hat session at TM Forum Innovate Americas 2026.

The future of network operations is still automated, but it's also more intelligent, enabling better decisions, faster MTTR, and a practical path from Level 4 toward Level 5.