- TM Forum reports progress toward Level 4 autonomous networks, moving past years of gradual "tinkering at the edges"
- Agentic AI is the primary engine of change, enabling proactive, multi-step reasoning and cognitive self-adaptation
- Early Level 4 adopters like China Mobile report steep cuts in maintenance costs and fault repair times
After years of "tinkering at the edges," the telecommunications industry has reached a point of "significant change" in the transition to Level 4 autonomous networks, fueled by a sudden surge in agentic AI, according to a recent report from TM Forum.
"For some time now the slow but steady transition towards autonomous networks has felt a bit like tinkering at the edges: gradual progress was being made but the big leap forward was yet to happen," according to the March 2026 report, "Assessing CSPs' progress toward Level 4 autonomous networks."
"But the second half of 2025, and the early part of 2026, have marked significant change. More and more operators are declaring, and validating, their attainment of Level 4 autonomy in specific network domains," according to the report.
TM Forum presented 37 certifications in November for progress toward Level 4 autonomous networks (AN), which the report explains is "the point at which networks represent a major shift from traditional human-defined automation processes to true autonomous decision-making."
What are the levels of autonomous networks?
TM Forum defines six autonomous network levels, numbered 0-5, with 0 denoting manual operations and maintenance, with assisted monitoring capabilities, and a rating of 5 used to designate fully autonomous networks. Level 4 is a "highly autonomous network" — not quite fully autonomous, but well on the way.
"Today, most operators are between Levels 2 and 3, but some are moving rapidly towards Level 4 in certain domains," according to the report.
Some 21% of respondents to a TM Forum survey said their companies are operating at Level 3 or above overall, compared with 19% last year. TM Forum surveyed 125 individual respondents in 80 companies worldwide.
Agentic AI is the engine of change
The primary engine behind this sudden leap from "tinkering" to Level 4 is the transition from traditional automation to agentic AI. Unlike basic generative AI, which some 32% of CSPs have already introduced into network operations, agentic AI employs proactive, multi-step reasoning to independently make and execute decisions with little to no human intervention. This "intelligence" layer is what allows a network to move beyond simple pre-configured rules to what TM Forum calls "cognitive self-adaptation," effectively giving the network a brain to manage the distributed, uplink-intensive data flows driven by 5G-Advanced and AI workloads.
Mobile RAN is the top priority for autonomous operations, followed by the IP network, the report says. "The highest-priority processes in those domains are those that improve customer experience (such as fault management, service assurance and service delivery)."
The mobile network makes sense as a starting point because of its complexity and the potential for saving energy.
But there's a qualification to these findings — TM Forum found that its testing criteria, the AN Level Evaluation Tool (ANLET), is excessively subjective and high-level. The group is working on enhancements, according to the report.
The productivity payoff is clear, according to the TM Forum report
Cost-cutting is a primary motivation for autonomy. "Given that most CSPs face significant challenges to growing revenues, reducing cost is an essential element to maintaining profitability, with savings often enabling further AN investment. Automated fault detection and resolution is the next highest priority, with CSPs keen to reduce the time-consuming nature of fault identification and negative impact of network faults on customer experience," according to the report.
"With high enough levels of automation, some CSPs estimate they can resolve 95% of trouble tickets without human intervention, in turn greatly reducing the cost of O&M. China Mobile, for example, already cites a more than 30% reduction in backend O&M manpower, as well as a 30% reduction in fault and customer complaint mean time to repair (MTTR) on average, by reaching Level 4 in its network operations centers," according to the report.
And Telkomsel reduced wireless traffic loss by 12.6%, improved MTTR by 6% and increased its customer experience index by 14.7% in a trial in east Indonesia of AI-driven predictve network operations, the report says.
Asian telcos lead
"CSPs are making progress towards Level 4 autonomous networks, but most are doing so gradually," according to the report. "Our research shows that while the industry expects meaningful progress over the next decade, operators remain cautious in their investment strategies. They are focusing on near-term operational gains, but still face significant integration and architectural challenges. They are also looking for ways to improve how they assess their progress."
While the report highlights Asian telcos as leaders in autonomous networking, Western operators aren't standing still. T-Mobile has a vision for intent-based networking, "turning the network into essentially a giant agentic AI platform where everywhere, all the AI agents and every network element work together, collaboratively and adaptively, to optimize network performance,” T-Mobile CTO John Saw said at Mobile World Congress 2026.
During Winter Storm Fern in January, one of the largest winter storms to ever hit the United States, with more than 1 million customers losing power, T-Mobile used its autonomous network to make around 30,000 antenna tweaks to keep as many customers connected as possible, Saw said.
Reality check
TM Forum paints an optimistic picture of telco automation. But a recent Accenture study was more pessimistic, finding 79% of telcos are still at Level 0/Level 1 autonomy, with only 22% expecting to reach Level 4 automation by 2030. Key obstacles include legacy BSS/OSS systems and limited talent, leaving many operators stuck in an incremental, hybrid approach to automation.
Download the report here: "Assessing CSPs' progress toward Level 4 autonomous networks."