Vladislav Zaimov is a seasoned authority in the telecommunications sector, specifically recognized for his deep expertise in enterprise infrastructure and the high-stakes world of risk management for vulnerable networks. As the industry moves past the initial wave of artificial intelligence hype, Zaimov’s focus has shifted toward the practical, often grueling work of integrating these technologies into live, mission-critical environments. Our discussion delves into the core themes of the recent Telco Days event held on November 6, where industry giants like Software Mind, Telefónica, IBM, and Amdocs gathered to chart a course for agentic AI. We explore the balance between autonomous systems and human oversight, the modernization of legacy foundations, and the emerging opportunities to transform traditional voice services into intelligent, context-aware platforms.
Moving AI from experimental labs into live production often faces hurdles like legacy infrastructure and organizational inertia. How are you seeing operators bridge the gap between these ambitious AI visions and the daily reality of network management?
The reality on the ground is that while the technology itself is moving at a breakneck pace, the organizations behind the networks are often struggling to keep up. During our discussions in Warsaw on November 6, it became clear that the primary blockers aren’t necessarily a lack of capable models, but rather a lack of internal readiness and the stifling weight of legacy systems. To bridge this gap, we are seeing a strategic shift where successful operators are moving away from implementing technology for its own sake and instead starting with very specific, measurable problems. For example, companies like Alianza and Next Mobile are beginning to focus on isolated network-automation tasks and specific fraud prevention protocols where the impact can be monitored closely. It is essential to remember that a massive, all-encompassing model isn’t always the right tool for every job; sometimes a smaller, more specialized model provides the response time and accuracy needed for live operations. By picking these targeted battles, operators can build the necessary confidence to eventually scale AI across their entire infrastructure without compromising the stability that their customers depend on.
With the rise of agentic AI, there is a growing interest in systems that can act autonomously rather than just providing data. What are the essential safeguards that must be in place before we let AI agents take the wheel in network operations?
Agentic AI represents a massive leap because we are moving from “AI as an advisor” to “AI as an actor,” which naturally introduces a new layer of risk regarding service availability and regulatory compliance. At Telco Days, the consensus among participants like ROCCO and Tallence was that we must implement a system of “constrained autonomy” where agents operate within very narrow, clearly defined permissions. Every single action an agent takes must be traceable and auditable, ensuring that if a fault occurs during a self-healing process, we can pinpoint exactly why a decision was made. I advocate for a gradual approach where these systems initially exist in an “observation mode,” gathering data and alerting human operators before they are granted the power to execute changes. This cautious progression is vital because an incorrect automated decision in a telecom environment doesn’t just result in a minor glitch; it can disconnect thousands of people and erode the fundamental trust that the industry is built upon.
The probabilistic nature of large language models often clashes with the deterministic requirements of telecommunications. How should operators decide which tasks are suitable for AI and which should remain governed by traditional rules?
This is a critical distinction that many people overlook when they get caught up in the excitement of Generative AI. We have to be disciplined enough to recognize that many telecom processes require predictable, repeatable, and auditable outcomes—things that traditional, rule-based systems handle perfectly. For straightforward decisions with a binary outcome, introducing a probabilistic AI model can actually add unnecessary complexity and risk. However, we see great success when these two worlds are combined; for instance, using a smaller AI model to interpret the nuances of customer language while leaving the actual execution of billing or service changes to a deterministic backend. This hybrid approach allows the AI to handle the “messy” human interaction part while the core network logic remains grounded in hard, predictable rules. It’s about using the right tool for the right job, and as we often say, sometimes the most sophisticated solution involves no AI at all for the most critical path.
Fraud management has become a sophisticated arms race, with generative AI being used to create more convincing scams. How is the industry evolving its defense mechanisms to protect subscribers in this new landscape?
The threat landscape has shifted dramatically because generative AI allows fraudsters to craft messages and voice scams that are nearly indistinguishable from legitimate communications. To counter this, we are looking at integrated signaling-security solutions, such as those found in the Amplitiv portfolio, which combine traditional signaling controls with advanced AI-based text analysis. This dual-layered defense is necessary because modern threats evolve much faster than any human team can manually update a rule set. By using AI to identify emerging patterns and suspicious messaging activity in real-time, we can flag threats before they reach the subscriber. However, the human element remains the ultimate safety net; we use AI to perform the heavy lifting of initial analysis on unfamiliar traffic, but we still empower human analysts to make the final call on how to handle the most complex or sensitive threats. This collaboration ensures that we stay ahead of the scammers without accidentally blocking legitimate traffic and frustrating our users.
There is significant buzz around the potential for AI to revitalize traditional voice services. Can you elaborate on how transcription and summarization could turn a standard phone call into a high-value service?
For too long, the industry has treated the network as a “dumb pipe” for connecting calls, but AI gives us the opportunity to treat the voice path as a rich source of context and data. By integrating tools like the Amplitiv AI Media Server directly into the voice path, operators can offer real-time transcription and call summaries that provide immediate value to both enterprise and retail customers. The beauty of this approach is that the processing happens within the operator’s secure environment, ensuring that sensitive conversational data never has to leave the network’s perimeter. Imagine a business call where a summary is automatically generated and synced with a CRM system the moment the call ends—that is a massive efficiency gain. By developing these additional services around voice connectivity, operators can move beyond simple price-per-minute models and provide an intelligent platform that makes every conversation more productive.
Modernizing the foundations of a telecom operator is notoriously difficult due to data fragmentation across billing and customer management systems. How do you recommend tackling this without disrupting current services?
Fragmentation is the silent killer of AI initiatives because these models are only as good as the data they can access, and currently, that data is often trapped in isolated batch processes. You cannot simply rip and replace these legacy platforms overnight without risking a catastrophic service failure, so the modernization must be a surgical, gradual process. Operators need to move toward a unified, real-time data environment where information flows seamlessly between billing, network operations, and customer support. This involves creating integration layers that allow new AI capabilities to “plug in” to existing systems while the core legacy infrastructure continues to run the day-to-day business. It is a transition of degrees, where you slowly offload specific functions to more modern, AI-ready architectures, ensuring that the foundation is solid before you try to build the more ambitious “agentic” layers on top.
What is your forecast for the role of AI in telecommunications over the next few years?
I expect that by the time we reach the late 2020s, the distinction between “AI” and “standard network operations” will have almost entirely vanished because intelligence will be baked into every layer of the stack. We will move away from the current era of experimental silos and toward a reality where self-healing networks are the baseline, managed by agents that act with high precision under strict human-defined guardrails. The focus will shift from simply “deploying AI” to “orchestrating AI,” where the primary skill for telco engineers will be managing the interactions between various specialized models. We will also see a massive resurgence in the value of voice services, as they become the primary interface for AI-driven personal assistants that live within the operator’s cloud. Ultimately, the winners in this space will be the companies that prioritize data integrity and operational safety today, as those will be the only organizations trusted to handle the fully autonomous networks of the near future.
