For Communications Service Providers (CSPs), the opportunity in Autonomous Networks extends beyond automating individual tasks. It is about improving reliability, accelerating service innovation and resolving problems before customers experience disruption.
Yet many operational processes still depend on human handoffs: an alarm triggers an investigation, a ticket moves between teams, an engineer acts and someone else checks the result. While the technical fix might be straightforward, coordination can delay recovery.
The CSP’s challenge is turning intelligent decisions into controlled actions and verifying service recovery and making sure that CSP can trust the autonomy process. This is where Agentic AI can help close that gap and provide evidence of success.
Is human expertise still required?
Agentic AI introduces reasoning, planning and more adaptive decision-making alongside established automation. Agents can examine network conditions, investigate likely causes, recommend remediation and interact with existing operational systems.
Specialised agents can collaborate across assurance, optimization and other functions, working towards objectives set by humans. Feedback from actions and outcomes can inform future investigations.
For CSPs, the opportunity is to connect the stages of an operational process more effectively through Agentic AI. Human expertise will remain essential in setting intent, defining policy and determining where autonomous execution is appropriate.
Closing the loop with evidence
To build complete autonomy, 5 pillars/stages are critical. These are: detect, reason, decide, act and assure, based on TMForum definition.
- Detection brings performance, fault, topology and other information together.
- Reasoning uses learned baselines and anomaly detection to distinguish degradation from routine variation.
- Decision-making builds an investigation around the specific trigger and produces an explainable diagnosis.
- Action connects that diagnosis to operational processes, including ticket enrichment, dispatch, customer notification and remediation. Rollback provides a response when a change does not produce the intended result.
- The final step of assurance is critical to show that customer experience has recovered. If recovery is incomplete, the investigation should continue.
As an example, a closed trouble ticket alone is not sufficient evidence of success. Trusted autonomy requires confirmation that the underlying service problem has been addressed.
Are agent recommendations the final verdict?
An agent’s recommendation does not automatically mean execution. It is important to define what an autonomous capability may do, which resources it may affect and when human approval is required.
Those boundaries become more important as autonomy expands across domains. A network optimization recommendation, for example, may need to account for service commitments (SLAs), congestion, maintenance and competing objectives.
Existing orchestration, controllers and network interfaces remain key execution mechanisms for remediation. Agents can determine an appropriate response while proven operational capabilities carry it out within the defined policy. Audit trails, observability and evaluation help operations teams understand what happened, why it happened and whether to continue, roll back or escalate. Humans will provide the final approval for execution of each recommended action (This is to achieve Autonomous networks L4).
When to scale autonomy?
A practical starting point, for realizing and scaling Autonomy with the help of Agentic AI tools, is one or two high-value scenarios (fractional autonomy), such as mobile network fault management or a fixed broadband complaint handling. Using the autonomy steps defined above (detect, reason, decide, act and assure), successful execution at each stage can provide a stronger foundation for the next expansion.
To scale up autonomy, acceptance of agent recommendation, accuracy of investigation and confirmation of recovery need to be in place. Once the CSP establishes that the fractional autonomy is providing effective business outcomes (such as Time to identify and resolve issues, intervention costs and customer impact) for the first few scenarios, it will build CSP confidence to expand autonomy across all domains. Annual operational saving is a good measure to track the success of fractional autonomy rollouts.
This thought leadership blog is a summary of Mycom’s webinar, Rethinking network operations: harnessing Agentic AI to deliver Autonomous Networks Level 4, featuring perspectives from Mycom, Omdia and AWS.
