Telecom’s AI revolution: from whitepapers to real-world autonomy, industry races toward self-driving networks

By mid-2026, the autonomous network industry has coalesced around a collaborative framework emphasizing Layered Intelligence and Open Collaboration, as championed by leaders like Yang Chaobin. This approach advocates for implementing real-time autonomous closed loops within single domains while achieving global orchestration through cross-domain intelligence, a vision underscored by over 100 industry partners signing the Autonomous Networks Manifesto and approximately 30 CSPs releasing blueprints and launching practical initiatives. Such widespread commitment reflects a collective push toward scalable, interoperable autonomous network operations that transcend vendor and technology silos.

Vodafone, Google Cloud, and TM Forum have concretized this collaborative momentum by jointly publishing a whitepaper that sets forth unified frameworks and open standards designed to transition autonomous networks from experimental proofs-of-concept to production-ready, trusted closed-loop automation. This initiative aligns closely with TM Forum’s autonomous network maturity framework, which emphasizes context-aware, cross-domain decision-making over isolated automation tasks, thereby fostering a shared industry direction that integrates network, service, and customer data to overcome fragmented operational silos.

Central to enabling scalable and trustable autonomous networks is the development of unified agent communication standards, notably the open-source A2A-T interface protocol, which facilitates seamless multi-agent collaboration across diverse network generations and vendors. This standardization effort, advocated by Yang Chaobin and realized through TM Forum’s Moonshot Catalyst program in partnership with Verizon and Google Cloud, underpins practical deployments like the industry’s first temporal digital twin—an AI-driven framework that exemplifies how interoperability and trusted AI can transform human-guided operations into autonomous network realities.

Operators adopting AI-driven unified intelligence frameworks that integrate telemetry, packet data, service intelligence, and customer experience into a 360º observability model have demonstrated tangible operational gains, including a 30% reduction in alarm noise and up to 80% faster troubleshooting cycles. These improvements, reported by Tier-1 CSPs in the US and Europe, validate the practical benefits of collaborative standards and shared operational views, reinforcing the industry’s confidence in open, unified frameworks as the foundation for trusted, scalable autonomous network automation.

Nokia and Databricks have pioneered a unified, cloud-agnostic data platform that addresses the telecom industry's fragmented data landscape by enabling AI-driven autonomous networks to scale real-time analytics seamlessly across multiple cloud and open-source environments. Their innovative approach uses platform-independent Python expressions and a custom compiler to translate abstract data workflows into native formats like Delta Live Tables and Flink SQL, eliminating vendor lock-in and accelerating deployment without manual rework. This architecture supports AI agents that autonomously generate new data products from natural language prompts, request human validation, and deploy pipelines automatically, exemplifying agentic AI’s transformative role in network automation innovation.

Nokia's migration of its full telecom operations stack to AWS marks a significant leap toward Level 4 network autonomy, integrating orchestration, assurance, unified inventory, and AI-driven closed-loop automation into a single cloud-hosted platform. Leveraging multi-agent AI systems on Amazon Bedrock AgentCore, Nokia autonomously optimizes network configurations by fusing real-time data with historical performance and operator intent, achieving automation rates exceeding 90% and reducing network slice rollout times by up to 85%. These operational gains, including service delivery under four hours and halving customer-impacting incidents, demonstrate scalable closed-loop automation moving decisively beyond proof-of-concept stages.

Verizon’s AI-driven roadmap exemplifies the industry’s shift from traditional network management to cognitive Level 4 automation, where generative AI and autonomous software agents execute over 70 million network configuration changes annually, drastically reducing manual effort and enabling engineers to focus on complex challenges. By integrating large language models into workflows, engineers specify outcomes via natural language prompts, fostering shared knowledge and operational agility. Their autonomous agents continuously monitor, diagnose, and remediate faults in production, cutting resolution times from hours to under two minutes and enhancing user experience through near real-time performance tuning using detailed operational and indoor location data.

Vodafone and Google Cloud’s development of the industry’s first temporal digital twin, in collaboration with TM Forum, offers a pragmatic blueprint for closed-loop network operations that build trust in autonomous networks by combining AI, real-time intelligence, and open industry frameworks. This temporal digital twin enables operators to transition from human-guided decision-making to scalable, business-aware autonomous operations, emphasizing interoperability, standards, and trusted AI as critical enablers. This shift mirrors trends in manufacturing where digital twins evolve from static simulations to active, operational tools managed by human coordinators overseeing teams of AI agents, signaling a broader transformation in how networks and systems are optimized in real time.

Telecom operators face formidable operational challenges in adopting agentic AI within mission-critical environments, primarily due to concerns over data quality, legacy system integration, and technical debt. Companies like RADCOM Ltd. and Kaya Global Inc. are pioneering cloud-native AI platforms and deterministic workflows to mitigate risks such as AI hallucinations, yet the industry remains cautious, prioritizing risk management and data readiness before scaling autonomous deployments. This cautious stance is underscored by the complexity of integrating sprawling legacy infrastructures with modern AI-driven architectures, as many operators grapple with siloed data and fragmented systems that hinder trustworthy AI insights.

Balancing the transformative potential of generative and agentic AI with operational governance remains an unresolved puzzle for telecoms, as the sector wrestles with the 'human in the loop' dilemma. While AI accelerates anomaly detection and predictive maintenance—benefiting 81% of operators surveyed by TCS—human expertise continues to be indispensable for interpreting AI outputs within network and business contexts to ensure secure, reliable decision-making. This hybrid approach reflects a broader industry consensus that AI autonomy must be carefully managed, with engineers maintaining oversight to prevent costly missteps in mission-critical scenarios.

Ensuring production readiness of AI agents extends far beyond technical capability, demanding robust governance frameworks that encompass trusted data, domain expertise, and stringent security guardrails. Experts like Peter Garraghan of Mindgard emphasize the necessity of resilient defenses against adaptive adversarial attacks, while Chih-Han Yu of Appier highlights that true readiness means AI agents must deliver measurable business value, not just technical accuracy. Moreover, continuous management—setting goals, reviewing performance, and maintaining clear escalation paths—is vital to sustain consistent outcomes, as Hemant Kashyap from Kindsight notes, underscoring that autonomy without oversight risks operational reliability.

Operational risks also emerge from AI agents’ potential to be manipulated or to act beyond intended parameters, necessitating rigorous identity governance and transparent permission controls before granting production-level autonomy. As Sanjay Dhawan of SymphonyAI warns, the critical question is not whether an AI can perform a task, but whether it can be coerced into unintended actions by input content. Additionally, cost overruns are a growing concern, with Praful Saklani of Pramata reporting AI agent expenses often exceeding budgets by two to three times due to complex autonomous reasoning. Finally, the ability of AI agents to recognize uncertainty and seek clarifications—rather than confidently making erroneous decisions—is crucial to reducing operational risk, a point stressed by Irina Bukatik of Branch.

Verizon’s AI-driven roadmap exemplifies a strategic leap toward Level 4 network autonomy by embedding generative AI and autonomous software agents that enable cognitive automation capable of independently resolving operational issues. Engineers now leverage advanced language models to specify network behaviors through natural language prompts, streamlining collaboration and accelerating fault resolution from hours to under two minutes, thus significantly minimizing customer disruption and operational overhead.

The $4 billion joint venture between BT and Verizon marks a pivotal industry move to address the escalating demand for AI-ready connectivity on a global scale, targeting over 3,000 multinational customers across 180 countries. This partnership not only accelerates next-generation, secure, and resilient network infrastructure deployment but also strategically allows both companies to focus domestically while leveraging combined international wireline assets to meet complex data sovereignty and regulatory requirements.

Agentic AI is redefining telecom growth paradigms by enabling operators to scale out to vast new user bases—including potentially hundreds of billions of intelligent software agents—scale up through personalized service models, and scale fast by drastically reducing time-to-market for new offerings. Huawei’s examples, such as a Thailand operator launching a football streaming package within a week and gaining 400,000 subscribers in three months, underscore how agentic operations simplify product complexity while boosting customer acquisition and satisfaction.

The evolution of network infrastructure toward a bifurcated model—combining massive centralized AI data centers with highly distributed edge computing—is critical to supporting emerging AI workloads and ultra-low latency applications like autonomous vehicles. However, as Uber’s Vishnu Acharya highlights, network connectivity remains a bottleneck, necessitating high-bandwidth links both within data centers and at the edge to enable scalable autonomous network operations, a challenge that aligns with Verizon’s emphasis on open standards and multi-vendor AI orchestration ahead of 6G deployment.