Autonomous networking is easy to claim and hard to prove. For Axiata, it is not a slogan, but a multi-year transformation program known as A3: Axiata’s Autonomous AI Network strategy.

A3 is designed to move the group beyond incremental automation and toward fully programmable networks. Built on cloud-native principles, open APIs and TM Forum’s Open Digital Architecture (ODA), the programme spans RAN, Core, Transport and Edge domains. It aims to decouple hardware and software, embrace microservices, introduce digital twins and create an infrastructure that can be managed by intent rather than manual configuration.

The ambition is structural change: lower operating and capital costs, faster service innovation, greater vendor flexibility and a response to structural pressures such as declining ARPU.

But ambition alone is not the measure.

As Dr. Tomasz Gerszberg, CTO at Axiata, explained, the journey is deeper than adding automation layers to legacy systems. Today, many networks operate with partial automation; what he described as “level 1.6.” The goal is Level 4 autonomy: a state where human operators define intent and the network executes it across domains with minimal manual intervention.

In simple terms, Level 4 means taking people out of routine operational loops and allowing systems to translate business intent directly into network action.

Measuring autonomy: Certification and ROI

Rather than debating maturity claims, Axiata has chosen a proof-based approach.

“How do we measure it? Very simple, the number of certified use cases by the TM Forum at the end of the year,” said Gerszberg. “No compromise. Just get the certification and prove.”

But certification is only half the equation. “Every use case must have positive ROI.”

That commercial discipline shapes the entire program. Automation is not pursued for technical elegance. It must pay for itself.

One lever is reducing reliance on external managed services. “One of the reasons how we can also deliver the ROI is just to internalize those services through automation,” he explained.

For operators wondering where to begin, Gerszberg advised starting pragmatically. “There are still the use cases that are super easy. Start with energy optimization. [It] delivers a lot of value.”

In markets facing rising energy costs, those savings alone can justify early automation investment.

Beyond cost: Autonomy as a revenue lever

The conversation also challenged the assumption that autonomy is purely a cost play.

In prepaid-heavy markets, where Axiata operates with up to 95% prepaid customers, network performance is directly tied to revenue. Every incremental increase in traffic, captured without additional capital investment, translates quickly into measurable gains.

“We are in the markets [with] 95% prepaid customers. Every growth of the traffic, without spending additional investments, gives very simple ROI,” Gerszberg noted.

“It’s a strong promise…not everyone would like to dare to give this promise to the CFOs. We take the risk.”

Data fabric: Why real-time autonomy changes the data equation

If Level 4 is the goal, the hardest problem is not orchestration. It is data.

Most enterprise data platforms are designed for analytics: historical insights, dashboards and reporting. Autonomous networks require something different; real-time, high-volume, short-lived data processing.

For network automation, data must be processed instantly. Information that is even ten minutes old may already be irrelevant. Gerszberg described this as “disposable data” or data that is valuable only in the moment.

Existing data architectures, particularly cloud-only models, are not always built for this kind of workload. Moving large volumes of time-sensitive network data back and forth to public cloud environments can introduce cost and latency challenges.

For Axiata, engineering and deploying a new data fabric that supports both cloud and on-prem environments is a priority for the year ahead.

Agentic AI: Why restraint may be strategic

Agentic AI introduces another layer of complexity.

Using AI agents to support analytics or back-office processes is relatively straightforward. The challenge arises when agents operate directly on live networks.

“More agents, more problems,” said Gerszberg.

The idea that each vendor supplies its own autonomous agent may sound appealing but in practice, it risks creating overlapping logic, governance gaps and coordination conflicts. Building layers of agent coordinators and “super agents” to manage other agents quickly becomes unsustainable.

Instead, he argued for coherent intelligence, a unified “super brain” for the network, rather than “thousands of agent Smiths.”

Transparency is non-negotiable. “I don’t want to have an agent in my network that has… any secret knowledge not known to me.”

Orchestration: Managing conflict, not just workflows

As networks become more programmable, orchestration becomes less about workflow automation and more about conflict resolution.

Conflicts arise across business priorities, service demands and resource constraints. “The most important role of the orchestrator will be the conflict management,” Gerszberg said.

Simple scheduling rules are not enough. Autonomous systems must reconcile competing intents and optimise outcomes across technical and commercial dimensions. That capability — potentially including negotiation between agents — may become the defining feature of next-generation orchestrators.

Change management: A question of scale and opportunity

Automation often raises concerns about job displacement. In Axiata’s case, Gerszberg sees opportunity instead of resistance.

In one operating company managing 20,000 sites with just 180 technology staff, automation enables teams to move faster and focus on higher-value work. “No one will lose the job there,” he said.

Instead, automation accelerates deployment cycles and can deliver market advantage. While some natural uncertainty remains, “we are never sure…if this agent will be smart enough,” he does not see structural resistance to adoption.