- Telekom Malaysia (TM) says agentic AI now handle 65% of its order-to-service processes
- TM's Chief Network Officer says domain engineers, not outside developers, built the agents
- The initiatives form part of a broader shift toward AI-enabled operations and digital services
Telekom Malaysia, the country’s incumbent operator and its largest fixed-line broadband provider, is using AI to improve efficiency across its network operations but says AI-led revenue opportunities have yet to emerge.
“What we are seeing is that the effort that we are putting into AI is pretty much on improving efficiency. There are no real key use cases that bring in higher revenue in terms of adopting or bringing over AI to consumers or enterprises,” said Mohamed Tajul Sultan, Chief Network Officer at TM, in an exclusive interaction on the sidelines of the recently concluded FutureNet Asia 2026.
Even so, AI agents now handle a significant part of TM’s order-to-service processes, while the operator is also using AI to detect network anomalies and warn customers before service quality deteriorates. “About 65% of our processes in terms of order-to-service are run by an AI agent. The remaining 35% work, including work like sending a technician to install a fixed line, cannot be automated,” revealed Mohamed Tajul Sultan.
This has led to more proactive service delivery. Instead of simply assigning customers an installation date, TM contacts them to arrange a time that works for them. The service provider is also using AI to detect network anomalies and warn customers when their service quality is deteriorating, allowing them to schedule repairs before a fault starts impacting them. "Many customers find it a bit weird," he said, because customers have traditionally reported faults themselves.
Mohamed Tajul Sultan revealed that TM has organized network operations into three key buckets: plan to build, order to service, and trouble to restore. Mohamed Tajul Sultan said trouble-to-restore was where the company first started using AI about a year back after carrying out literature reviews and extensive sandbox testing.
For order-to-service, each customer order has to be orchestrated across several network layers, which used to be handled domain by domain and passed between teams. Agents now coordinate that work.
TM's strategy is in line with the wider pattern. Telcos across the world are using AI to move operations from reactive to predictive and to reduce the manual hand-offs. While the service providers are using AI to ensure more efficient network operations, it is not leading to an increase in revenue.
TM has devised the PWR 2030 strategy to become a digital powerhouse by 2030. It focuses on digitalization, infrastructure leadership and building Malaysia’s digital talent and innovation ecosystem. The company is also targeting to expand beyond core connectivity into digital services such as data centers, cloud, GPU-as-a-Service and AI, as part of its strategy.
How is Telekom Malaysia using AI for new revenue streams?
While the telcos continue to struggle for use cases, some service providers have started selling tokens, which may emerge as a viable business model. “In the next few years, the opportunity will increasingly be around combining connectivity with digital and AI services for consumers and enterprises. We are already seeing AI agents that can help small businesses with functions such as HR and analytics, capabilities they may not have the resources or headcount to manage themselves,” said Mohamed Tajul Sultan.
“Over the next three to five years, I expect AI to become increasingly accessible even to smaller businesses. This creates a significant opportunity for telcos to move beyond connectivity and deliver value-added AI and digital services. That is also where operators can begin generating stronger returns on the substantial investments they have made in their networks,” he added.
He believes that the coming few years will see telcos moving from being a connectivity operator to becoming a digital company.
Engineers becoming developers
Interestingly, he revealed that TM did not hire developers to build the agents, because developers do not understand how network systems work. Instead, the company decided to train its own domain engineers to code. "I did not bring a single person from outside," he said.
Participation was voluntary, and he said many engineers volunteered or taught themselves, helped by competition among thousands of engineers and the fact that coding is now far easier than it was. He said the main obstacle at the start was a mismatch: he had developers who were not engineers, and engineers who were not developers.
TM builds its own agents rather than asking vendors to build them, which he said has largely removed the interoperability problem that has long dogged telcos. Vendors still share information to smooth integration. The company builds agents in small teams so that knowledge survives staff turnover, and it runs a governance framework and policy review before any agent is deployed.
Significantly, TM’s network department has not reduced any staff because of adoption of AI-led automation. "Until today, we have never reduced any staff."
Preparing the network for AI traffic
As AI and inference traffic continues to surge, service providers are struggling to grow network capacity to address this. Commenting on this, Tajul said capacity is only one part of the answer.
“Network readiness for AI is about more than having enough capacity. Latency is equally critical because users expect information almost instantly. Operators need to understand their capacity and latency requirements and how the network, cloud, compute, storage and data centers work together. That interoperability is extremely important, which is why we use sandbox environments to test whether capacity, latency, compute and storage are sufficient,” he elaborated.
Read more about agentic AI on Fierce Network
Vultr and Vast extend partnership to meet AI infrastructure demand
AI agent governance is having its OpenStack moment