Just a few years ago, artificial intelligence in telecoms networks was mainly associated with pilot schemes and proofs of concept – limited roll-outs tested on selected parts of the network, often without having much impact on the day-to-day work of operational teams. However, industry reports published at the turn of 2025 and 2026 show a clear shift: the sector is beginning to move away from experimentation towards live deployments and measurable business outcomes.
What the latest reports say
In its Tech Trends 2026 report, Deloitte points out that organisations must move from experimenting with AI to achieving concrete business results. Simply launching a pilot is no longer enough – the question of whether a new solution actually improves efficiency, reduces costs or streamlines operational processes is becoming increasingly important.
At the same time, Deloitte cites a Gartner forecast suggesting that over 40 per cent of projects based on agentic AI could be cancelled by the end of 2027. The problem need not lie with the technology itself, but rather with how it is implemented – particularly attempts to impose new AI mechanisms on processes and architectures that were not designed with autonomous agents in mind.
In the telecoms sector, the scale of investment is also growing rapidly. According to a forecast by Mordor Intelligence, the agentic AI market in telecoms and network management is set to grow from around US$3.75 billion in 2025 to US$4.63 billion in 2026, and reach approximately US$13.35 billion by 2031.
This does not yet signify a widespread transition to fully autonomous networks. However, it does show that agentic AI is increasingly featuring in the strategies of operators and technology providers as a tool to support analysis, automation and decision-making.
Networks designed with AI in mind
These changes extend not only to how existing networks are managed, but also to the direction of development for future architectures.
The International Telecommunication Union is working on IMT-2030, the framework for the future generation of 6G networks. The scenarios being developed include, amongst others, Artificial Intelligence and Communication, Integrated Sensing and Communication, and the concept of Ubiquitous Intelligence.
Future networks are therefore intended not only to transmit data between users and devices, but also to support distributed processing, machine learning, sensor technology and decision-making closer to where the data is generated.
This direction is in line with the industry’s evolving concept of AI-native networks – networks in which AI mechanisms are designed as an integral part of the architecture and operational processes, rather than merely as an additional layer superimposed on the existing environment.
The integration barrier – still a challenge
Despite the progress made, one of the greatest obstacles remains the integration of new AI mechanisms with existing OSS/BSS systems.
Telecommunications networks are complex, multi-vendor environments that have often been developed over decades. Data relating to infrastructure, services, customers, and configurations may be stored across numerous independent systems and databases.
It is precisely this fragmentation of information that becomes one of the most significant constraints when scaling up AI.
An algorithm can only make accurate decisions if it has access to up-to-date, consistent, and properly structured data. If the view of the infrastructure is incomplete, out of date, or scattered across different systems, even an advanced AI model will operate on a limited information base.
Therefore, implementing AI on a larger scale often requires, first and foremost, organising the data and ensuring a consistent view of resources, services and dependencies within the network.
Automation also requires trust
The second key factor is trust in the decisions made by autonomous systems.
For years, operators have been cautious about handing over control of processes relating to critical infrastructure to algorithms. Concerns centre primarily on the ability to explain the system’s decisions, monitor its actions and intervene swiftly in the event of abnormal behaviour.
Consequently, the development of agentic AI is increasingly accompanied by mechanisms for oversight, control and auditability.
The network can automatically detect problems, analyse their causes and initiate corrective actions, but this does not mean relinquishing control on the part of the operator. In practice, mature automation should combine system autonomy with clearly defined operating principles, the ability to trace decisions and a human role where it is needed.
Where does SunVizion fit into all this?
This is precisely why we are developing SunVizion in a direction that combines detailed knowledge of the infrastructure with intelligent automation.
The AI Network Planner supports the automatic design and optimisation of networks, whilst the AI assistant enables teams to work with data using natural language, without having to manually search for information across different system views.
However, the data layer is of key importance.
AI can only effectively support design, analysis, or operations if it has access to reliable information about assets, their location, configuration, and interdependencies. Network Inventory thus becomes not only a system for documenting the infrastructure, but also a data source for increasingly advanced automation mechanisms.
In such a model, AI can take over some of the repetitive tasks, leaving the engineer to make decisions that require domain knowledge, experience and responsibility.
Operators who are already investing in the quality and consistency of their network data will be in a far better position to scale up AI than organisations which first implement advanced algorithms and only later begin to organise information about their infrastructure.
