Following Telco Days 2026, Software Mind’s annual conference focused on the practical implications of agentic artificial intelligence and network security, the discussion among telecom industry participants is increasingly centered on a fundamental question: how can operators move AI beyond experimentation and deploy it safely in live environments?
Held in Warsaw and online on November 6 under the theme “Roadmap to Agentic AI and Safer Networks,” Telco Days brought together representatives from companies including Software Mind, Amdocs, Alianza, IBM, Next Mobile, Telefónica and Tallence, alongside analyst firm ROCCO. Its agenda examined AI applications in telecom operations, fraud management and network protection, with an emphasis on real-world implementations and operational challenges.
In a follow-up interview after the event, Dawid Mielnik, telecom general manager at Software Mind, discussed the main issues surrounding the industry’s adoption of AI, including the emergence of agentic systems, the importance of human oversight, and the opportunities for operators to develop new services.
One of the central challenges, Mielnik said, is that moving AI into production requires more than access to increasingly capable models. Data privacy, security, legacy infrastructure, and organizational readiness can all complicate deployment. “The blockers are not because the technology is not there, but because the organization is somewhat not ready,” Mielnik said.
For operators, this means that successful AI initiatives need to begin with clearly defined problems and measurable objectives rather than technology for its own sake. Mielnik recommended focusing on specific applications, including selected network-automation tasks, fraud prevention, and customer-service processes.
The choice of model also matters. A larger model is not necessarily the best option for every application, he explained. Operators should select technology according to the task, taking into account requirements such as response time, accuracy, and operational complexity.
Agentic AI requires operational safeguards
Agentic AI was a central theme of Telco Days 2026, reflecting the industry’s interest in moving beyond systems that simply analyze information or generate recommendations.
Mielnik said early applications are emerging in areas such as network fault detection, limited self-healing processes and customer care. Examples include helping customers understand billing issues and supporting service-related interactions.
However, expanding these applications requires more than connecting an AI agent to operational systems. Mielnik emphasized the importance of reliable, accessible data and appropriate controls over what agents are allowed to do.
In particular, agents should operate with clearly defined and limited permissions. Their actions need to be traceable, and operators must be able to monitor and audit decisions, especially when those decisions can affect network operations or customer services.
Rather than granting agents broad autonomy immediately, Mielnik described a gradual approach. Systems can initially observe activity, gather additional information, and alert human operators before being allowed to carry out increasingly autonomous tasks. This is particularly important in telecommunications, where an incorrect decision can affect service availability, customer trust and regulatory obligations.
Keeping human oversight in the decision process
The need for safeguards is also connected to the nature of AI models themselves. Large language models generate probabilistic responses, while many telecom processes require predictable and auditable outcomes.
Mielnik said operators should distinguish between tasks that benefit from AI and those that are better handled through conventional rules and deterministic systems. Straightforward decisions with clearly defined outcomes may not require an AI model at all.
Other applications can combine technologies. A smaller model, for example, may help interpret customer language, while more complex analysis can take place outside the immediate service path, where response times are less critical.
The same principle applies to fraud management, another major subject at Telco Days. Generative AI is making fraudulent messages and other scams more convincing, increasing the challenge for operators seeking to protect subscribers and their networks.
Software Mind’s Amplitiv portfolio includes signaling-security solutions. The company describes the use of signaling controls and AI-based text analysis to help identify suspicious messaging activity.
Mielnik discussed combining established rules with AI-based analysis to address threats that may evolve faster than manually maintained rule sets. In the approach he described, AI can help identify potentially fraudulent messages, while human analysts retain responsibility for deciding how suspicious activity should be handled.
He also discussed the possibility of using generative AI to assist with the initial analysis of unfamiliar traffic patterns and potential threats. Under that approach, human operators would retain the final decision-making role.
Voice services offer another AI opportunity
Beyond network operations and security, Mielnik sees potential for AI to create new services around operators’ existing voice infrastructure. Telecommunications companies already carry large volumes of voice traffic. AI could enable additional services by making conversations easier to process, analyze, and use as a source of context.
One example is the use of AI to transcribe calls and generate summaries. Software Mind’s Amplitiv AI Media Server is designed to support AI capabilities in the voice path, including transcription and call summaries. The company describes an architecture in which call-related AI processing and generated content can remain within the operator’s environment.
During the interview, Mielnik described a demonstration involving an in-network voice call that was transcribed and summarized using AI. Such capabilities could provide information that other services can use, while allowing operators to maintain control over the underlying voice service and its associated data.
The opportunity, he suggested, is to develop additional services around voice connectivity rather than treating the network solely as a means of connecting calls.
Modernizing the foundations
Realizing these opportunities also depends on addressing the limitations of legacy telecom environments. Mielnik said information is often distributed across network, billing and customer-management systems. In some cases, these systems exchange information through batch processes or point-to-point integrations rather than through a unified, real-time data environment.
That fragmentation can make it difficult to provide AI applications with the information they need. It also means operators cannot assume that legacy platforms can simply be removed and replaced.
Instead, modernization generally needs to take place gradually, with existing systems continuing to operate while new capabilities are introduced and integrated.
The broader message emerging from Telco Days and the subsequent discussion is that the telecom industry’s AI transition is no longer just about testing what the technology can do. Operators need to determine where AI can deliver practical value, how much autonomy to grant it and how to maintain control over decisions that affect networks and customers.
For Mielnik, the objective should be to apply the appropriate technology to each problem, rather than introducing AI simply because it is available. “Sometimes also the right amount of AI is no AI at all,” he said.
Table of Contents
