- Many large operators remain far earlier in network automation maturity than public claims suggest
- Bolt‑on AI layers can deliver short‑term gains, but risk prolonging deeper structural problems
- Agent‑to‑agent interaction could reshape automation — but governance and trust remain unresolved
FUTURENET WORLD 2026, LONDON – On the stages here at FutureNet World in London, operators seem confident in describing their progress toward autonomous networks. Level 3.4! Level 3.7! Level 4! Don't get too excited, though. There is some "spin doctoring" going on.
Privately, the picture looks very different, according to industry execs from Blue Planet who spoke with Fierce Network today.
“Publicly, they’ve probably bumped up where they are — somewhere between Level 3 and Level 4,” said Kailem Anderson, VP of global products and delivery at Blue Planet. “But behind the scenes, when we talk to them…there’s one extremely large carrier in North America that’s acknowledged they’re somewhere between Level 1 and Level 2.”
Thankfully, the gap between public claims and operational reality is beginning to narrow — not because automation has accelerated dramatically, but because operators are starting to admit where they actually stand.
The “true acknowledgement” of where operators are versus the “spin doctoring” is actually “a good thing,” said Anderson. “We need to have a reality check on what's going on."
The problem with claiming Level 4 autonomy for just one network function — such as power monitoring, for example — is that it defines network automation too narrowly and doesn’t include planning, fulfillment and assurance across the network, said Gabriele Di Piazza, VP of product management, alliances and architectures at Blue Planet.
“Level 4 is multi‑layer automation,” Di Piazza said. “If you’re automating one function, that’s not Level 4. That’s still a function you’re automating.”
Is automation measurement becoming outdated?
So, what does Level 4 network automation stand today?
According to TM Forum, “Level 4 NA introduces decision-making based on intent-driven, predictive analysis, and the capability to perform closed-loop management of service-driven and customer experience-driven networks via AI modeling and continuous learning.”
The forum developed a six-level network automation taxonomy that can help operators discover where they are on their journey in 2019. The 3GPP, ITU and NGMN Alliance also have autonomous networks taxonomies, which launched in the early 2020s. GSMA launched theirs in March.
Di Piazza questioned whether these types of long‑standing autonomy models are still accurate in an era of agentic AI.
“The levels were codified [almost] 10 years ago,” he said. “They’re still valid in theory, but now we’re building automation that can generate configurations, interpret intent and act autonomously. That introduces a new trust cycle.”
Historically, operators relied on models and machine learning systems that could be validated over time, he said, noting that agentic AI changes the equation. “Now you need to restart the level of trust,” he said, particularly when generative systems are capable of hallucination or compounding errors.
This all doesn’t mean agents aren’t viable — but it does mean autonomy may not progress in a neat, linear fashion, Di Piazza said.
Bolted-on automation vs. structural change
One major sticking point to network automation for operators is their legacy BSS/OSS. Vendors in the space are taking varied approaches to that sticking point. Should they layer automation over existing BSS/OSS stacks or undertake the rip-and-replace, which is expensive, time-consuming and risky.
“Operators aren’t good at ripping and replacing,” Anderson said. “They’re good at building on top of legacy.”
The problem with that approach is the data challenge that come with it. Bolted‑on AI layers often inherit fragmented data, siloed processes and closed architectures, noted Anderson. “Where are you pointing your agentic layer?” Anderson asked. “If the data is all over the place, the outcomes are only as good as what you can access.”
“You cannot just layer things on top,” said Di Piazza. Referring to early NFV missteps, he said that treating agentic AI as an add‑on risks delaying real progress. “This technology needs to be an architectural foundation for how software is built.”
Netcracker’s AVP of Strategy Susan White, told Fierce that if a telco isn’t modernizing its BSS/OSS, then at some point, they’re “going to get stuck.”
Stuck because of the challenges around extracting data, but also because that data has to be correct and updated, she said.
“If you haven't updated, if you've got lots of inventory systems and they're not all up to date and all out of sync with each other, then you're giving the agents incorrect information,” White said. “You're never going to progress agentic AI in your company.”
The bolt-on approach, such as Amdocs' aOS agentic AI operating system, which is designed to ride on top of the traditional telco OSS/BSS stack, is necessary in some situations, according to Roy Chua, principal at AvidThink. "Because for existing customers, they already have the OSS/BSS running. They don't want to touch anything. It's the path to least resistance."
This method delivers value without destroying whatever they've already done, noted Chua. "I think that also, simultaneously, it gives them time to improve and swap out pieces of it. Because the bottom line is that they also have modules that they've written for customers. And they work. You don't want to mess with that."
The bottom network automation line
If there is one big takeaway from FutureNet World in London, it's that automation in operator networks is real but uneven, and held back by legacy systems.
In fact, during a panel today, Cathal Kennedy, Group CTO (acting) at Telenor, revealed that his pie-in-the-sky autonomous network goal is to be a network that operates like a fully automated car manufacturing plant that only employs 20 people. That dream will likely remain out of reach for many operators for a long time.
First, because agentic AI may ultimately redefine how networks operate, it also resets the trust, governance and architectural challenges operators must solve. Second, telecom networks are a critical infrastructure that demands human oversight.
"A lot of what we do [in the telecom industry] is mission critical, which is why governments are involved in making sure, and especially with AI, that we can govern and control. We always need that human ability to oversee, and not just oversee, but change whenever we want.”
Could 20 people oversee a whole network? Chua said, "It is conceivable, and it should be possible if you look at data center networks. With the right tools; with the right foundations, you could run a nationwide network with 20 engineers ... and I think we will get there in the next three years."
For now, the industry’s biggest goal may simply be admitting where it truly stands today so that it can get moving. 2029 or 2030 is not that far off. Let's go!