How the shift from Machine Learning to generative and now agentic AI is reshaping the very fabric of telecom operations, turning networks into intelligent autonomous systems. Welcome to part four of our “Engineering Smart Networks & Operations” mini-series.

Artificial intelligence has been part of the telecoms industry for decades. Long before the buzz about generative and agentic AI became part of the business lexicon, communications service providers (CSPs) were using machine learning to forecast demand, optimize networks, predict maintenance schedules and automate maintenance. In many ways, telecoms helped define industrial AI.

The last few years, however, have seen a profound acceleration. The sector has moved rapidly from machine learning to generative AI and now to agentic AI, systems conceived to act autonomously to achieve goals rather than simply responding to instructions. This is not an incremental upgrade; it is a structural transformation of how networks are built, operated, and monetized.

What makes this moment so important is the combination of maturity and momentum. Cloud-native architectures, specialized compute such as GPUs and NPUs, and the emergence of large language models have enabled the development of sophisticated platforms for AI that can operate at true network scale. Intelligence is no longer an overlay; it is becoming the organizing design principle of next-generation telecom infrastructure.

Two Arenas of Transformation

The most immediate opportunities for AI lie in two interconnected areas: autonomous networks and AI-driven operations.

Autonomous networks are the industry’s long-held vision of systems that are designed to detect, diagnose, and heal themselves, supporting engineers by freeing up their time so they can focus on innovation. This is where agentic AI begins to shine. It enables systems that can monitor network performance, isolate anomalies, and resolve issues before customers notice them.

AI-driven operations systems extend that intelligence across the enterprise. From network planning and supply-chain optimization to service design and configuration, AI advances are reshaping how telecoms deliver and sustain value. Well designed AI systems have the power to improve quality of service, reduce operational expenditure, and free scarce expertise for higher-order tasks, once the required training and quality of data are in place to ensure the technology can function as intended.

Imagine the change for the team at a national operator managing thousands of 5G cell sites. Instead of conducting manual analysis, their AI systems will continuously ingest network telemetry and customer-experience data to predict network congestion or degradation before it occurs. If a site shows unusual latency, the platform correlates the issue with equipment logs and environmental factors, then makes recommendations to an engineer (or, if appropriate, automatically performs) a configuration change to restore service. At the same time, generative AI updates the incident report, adjusts future maintenance schedules, and notifies the relevant engineers. What once required hours of investigation becomes an instant, closed-loop response.

Together, both areas address the twin imperatives driving every CSP: improved efficiency and productivity while unlocking new revenue. Self-healing networks lower downtime and maintenance costs, predictive analytics improves network utilization, energy efficiency and customer experience, and generative design tools allow operators to create personalized services that increase customer value. AI has become both a lever for efficiency and an engine for creativity in one of the world’s most capital-intensive industries.

The Challenge of Deployment

Given the considerable opportunities AI (and agentic AI in particular) offer the telecoms industry, why is it proving so hard for CSPs and network operators to make the leap forward?

Despite growing momentum, most CSPs continue to deploy AI in isolated pockets. A predictive maintenance pilot here, an AI-assisted chatbot there, each delivering value in isolation, but collectively falling short of coherent transformation. The real challenge is not experimentation, but industrialising AI across the value chain.

At Capgemini, we see that overcoming this challenge is about addressing three connected requirements.

The first barrier to scale lies in creating the solidfoundations required to unlock intelligence at scale. This means more than deploying models; it requires access to the right compute, platforms, and data.

Operators must make complex strategic choices. Should they invest in their own GPU infrastructure, partner with hyperscalers, or pursue sovereign AI models that keep sensitive data within national borders? Each option brings trade-offs across cost, performance, regulation, and control. At the same time, AI systems are only as powerful as the data they consume. Fragmented, low-trust, or delayed data keeps AI stuck in the pilot phase rather than embedded in operations.
 
Without a scalable AI-ready infrastructure and a unified data foundation, intelligence will remain episodic instead of systemic.

Even with the right technical foundations, AI cannot scale without organisational readiness. Telecom operators must adapt how they operate, govern, and deploy intelligence.

Historically separate domains, network, IT, and operations all need to converge around a shared data fabric and common operating model. Governance frameworks must also evolve to manage regulatory diversity, data sovereignty, and responsible AI requirements across regions. What works in one market may not be permissible in another.
 
AI readiness also depends on people and operating models. CSPs are not AI-native organisations, yet they now require skills that bridge deep network expertise with data science, automation, and transformation leadership. Without the right enablers, guardrails, and accountability, AI initiatives struggle to move beyond proof-of-concept.

Finally, true scale depends on how humans and AI work together. Agentic systems will increasingly act autonomously, but not entirely alone. Their impact is going to be shaped by trust, clarity, and design – all of which rely on the interaction between AI and humans.

Operators must intentionally define when AI decides, where humans intervene, and how accountability is maintained. Poorly designed interactions limit adoption; well-designed human-AI collaboration accelerates it. Just as team chemistry determines human performance, human-AI chemistry determines how deeply AI can integrate into telecom operations.

Accelerating the Transformation

Overcoming these challenges requires more than new tools. It demands a new way of thinking about how intelligence, infrastructure, and people evolve together.

By aligning these three requirements, telecom operators can move beyond fragmented experimentation and begin building AI-native networks that learn, act, and improve continuously, at scale, and with confidence.

There is already clear evidence of what this transformation can deliver across the telecoms industry. A European Tier-1 CSP, for example, has built a centralized big-data platform capable of processing more than 30 billion network events every day. By turning this data into real-time insights, the operator has achieved a step-change in network performance and significantly improved the efficiency of its network analysis.

Similar gains are being seen in operations. A North American cable operator has reimagined how its network operations center functions, deploying AI to correlate more than 16 million alarms in seconds. The result has been a 45% reduction in network incidents, alongside dramatic improvements in operational efficiency.

At the vendor level, the impact is equally compelling. A global network equipment provider has implemented an AI-enabled observability solution across its applications and networks, achieving over 80% proactive anomaly detection and 90% enriched performance insights. By embedding intelligence directly into observability, the organization has shifted from reactive monitoring to predictive, insight-led operations.

Together, these examples illustrate how AI, when applied at scale rather than in isolated pilots, is already delivering measurable performance, resilience, and efficiency gains across the telecoms value chain.

Through its work on these and many other, similar engagements, Capgemini has built a clear, experience-led view of what it really takes to make AI work in telecoms. Again and again, projects show that success is not driven by algorithms or tools alone, but by the ability to combine scalable AI capabilities with high-quality network data, embed them into operational reality, and design them in a way that engineers and operations teams can trust and use day to day.

Across customer programs, we have also seen how the hardest challenges are often less technical than expected. Aligning data sources across network and IT domains, putting the right operational guardrails in place, and ensuring that AI systems complement human expertise rather than override it are what determine whether initiatives scale or stall. When these foundations are in place, AI moves beyond experimentation and starts to deliver consistent, repeatable value.

What these examples collectively demonstrate is that generative and agentic AI are already reshaping the telecoms value chain when applied in this more holistic way. Reactive maintenance gives way to proactive intelligence, manual analysis is augmented by automated insight, and operations teams shift their focus from firefighting to orchestration. The result is networks that are run not just faster and more efficiently, but with greater confidence, resilience, and intent.

A Moment of Realistic Optimism

The telecoms industry is entering an era where networks will sense, learn, think and act for themselves. The evolution from machine learning to generative and now agentic AI marks one of the fastest and most far-reaching transformations in the industry’s history.

Yet progress will depend as much on pragmatism as ambition. AI in telecoms is not about replacing people or discarding legacy systems; it is about orchestrating intelligence across them. The organizations that succeed will be those that recognize the complexities, invest in trusted partnerships, and build the foundations for AI at scale.

With the right strategy and ecosystem, the industry can move beyond experimentation into a future of autonomous, adaptive, and resilient networks. For telecom leaders, this is not just an opportunity. It is an inflection point.