> **来源：[研报客](https://pc.yanbaoke.cn)** # CLOUD-NATIVE NEXT CHAPTER - AGENTIC AI-BASED OPERATING MODELS SUMMARY ## Executive Summary This document outlines the NGMN Alliance's vision for evolving cloud-native telecom operations by integrating advanced AI technologies. It builds on the NGMN Cloud-Native Manifesto and the Cloud-Native Computing Foundation (CNCF) Cloud-Native Maturity Model (CNMM), proposing a structured roadmap for mobile network operators (CSPs) to adopt GenAI and Agentic AI-based operating models. The framework defines five AI adoption levels, aligned with CNMM maturity levels, to guide operators through a transition from basic automation to fully autonomous, intelligent, and resilient network operations. The content highlights the importance of industry collaboration, responsible AI governance, skill development, and continuous optimization in achieving these goals. --- ## Core Content ### Cloud-Native Maturity Model (CNMM) Overview The CNMM is a structured framework that evaluates an organization's maturity in cloud-native adoption across **People, Process, and Technology** dimensions, and aligns it with **Business Outcomes**. It defines five maturity levels: 1. **Level 0 – Legacy**: No cloud-native adoption. 2. **Level 1 – Initial (Ad-hoc)**: Minimal understanding; unstructured adoption. 3. **Level 2 – Repeating (Opportunistic)**: Basic practices implemented inconsistently. 4. **Level 3 – Defined (Systematic)**: Standardised practices aligned with business goals. 5. **Level 4 – Managed (Measured)**: Proactive monitoring and optimisation. 6. **Level 5 – Optimised**: Continuous improvement and innovation. The CNMM serves as the foundation for integrating Agentic AI into telecom operations, enabling a structured transition to autonomous, AI-driven network management. --- ## Key Dimensions of CNMM - **People**: Skills, culture, leadership, security, and training. - **Process**: Workflows, governance, automation, and incident management. - **Technology**: Cloud-native platforms, architecture, integration, and security tools. - **Business Outcomes**: Cost optimisation, network efficiency, customer experience, time-to-market, and regulatory compliance. --- ## Agentic AI-Based Operating Models ### Maturity Mapping to CNMM Agentic AI-based operating models are aligned with CNMM maturity levels, and are categorized into five levels of AI adoption: #### **Level 1 – Foundational (Aware, Safe Pilots)** - **Purpose**: Safe exploration of GenAI with clear guardrails. - **CNMM baseline**: Level ≥2 across People/Process/Technology. - **Readiness gates**: Written policy approved, model card template adopted, PII scanning in place, pilot review checklist defined. - **Use Cases**: AI-assisted documentation, code notes, meeting minutes, ticket summarisation. #### **Level 2 – Enabled (Controlled Pilots, Initial Production)** - **Purpose**: Move from ad-hoc PoCs to controlled early production with basic LLMOps. - **CNMM baseline**: Level ≥3. - **Readiness gates**: Registry + signed artefacts, evaluation reports required, SLOs defined, basic runtime observability. - **Use Cases**: NOC copilot, RAG over SOPs, change request drafting, business reporting Q&A. #### **Level 3 – Integrated (Production LLMOps, SLO-backed Services)** - **Purpose**: Standardise LLMOps and integrate GenAI into operational workflows with SLOs. - **CNMM baseline**: Level 3-4. - **Readiness gates**: SLOs reliably met across two+ services, automated rollbacks, drift detection live, model governance board operational. - **Use Cases**: Customer care GenAI, network planning assistant, change risk assessment, closed-loop suggestions. #### **Level 4 – Closed-Loop (Autonomous-1, Safety-Gated Automation)** - **Purpose**: Introduce safe autonomy in well-bounded loops; scale multi-tenant, efficient inference/training. - **CNMM baseline**: Level ≥4. - **Readiness gates**: Documented set of closed loops with risk taxonomy, zero P1 safety incidents over N quarters, confidential pipelines validated, carbon/cost-aware scheduling in pilot. - **Use Cases**: Self-healing auto-remediation, proactive capacity scaling, digital twin-driven change validation, LLM-assisted policy recommendations. #### **Level 5 – Optimised (Responsible Autonomy at Scale)** - **Purpose**: Enterprise-wide, audited, sustainable GenAI with federated governance and continuous optimisation. - **CNMM baseline**: Level 5. - **Use Cases**: AI-optimised network slices, autonomous operations with human oversight, GenAI at scale for customers and internal users, AlaaS for partners. - **Outcomes**: Predictable ROI, audited compliance, continuous cost/CO2e reductions, high developer/analyst productivity. --- ## Tools and Frameworks Required To enable GenAI-based operating models, telecom operators need the following tools and frameworks: ### Foundational Components - **Data Architecture**: Robust and scalable infrastructure for structured and unstructured data. - **Digital Twin**: Virtual representations for simulation, testing, and validation. - **AI Flow Building Tools**: Platforms for creating, managing, and orchestrating AI workflows. - **Organisational Skillsets**: Cross-functional teams with AI/ML, data engineering, network operations, and AI governance expertise. ### Infrastructure and Platform - **Cloud-native platforms**: Containerisation, orchestration, and automation tools (e.g., Kubernetes). - **CI/CD pipelines**: Integration with GitOps for configuration and prompt management. - **Monitoring and observability**: Tools like Prometheus, Grafana, and OpenTelemetry for real-time insights and anomaly detection. - **Security tools**: Model artifact signing, policy-as-code for ML, confidential computing, and runtime protection. ### AI Model and Development Platforms - **Model serving**: KServe or equivalent. - **Training/evaluation**: Kubeflow, Ray. - **Batch/gang scheduling**: Kueue, Volcano. - **Service mesh**: mTLS for secure communication. - **Vector DB**: For internal RAG use. - **LLM observability**: OpenLLMetric, autoscaling (KEDA, HPA). --- ## Evolving Roles in Organisations - **People**: Shift in roles to include LLMOps, AI Ops Engineer, Prompt Engineer, and Network Data Scientist. Emphasis on Responsible AI literacy and Human-in-the-Loop (HITL) workflows. - **Process**: Redesign of workflows to include AI-driven automation, governance, and compliance checks. Use of synthetic traffic for continuous evaluation, and AI incident postmortems. - **Technology**: Integration of AI into cloud-native platforms, with enhanced security, observability, and resource management. - **Organisational Impact**: Change management, cross-industry collaboration, and alignment of business outcomes with AI capabilities. --- ## Key Success Factors - **Industry Collaboration**: Essential for standardisation and innovation. - **Responsible AI Governance**: Ensuring data governance, security, and compliance. - **Skill Development**: Building cross-functional expertise in AI, cloud-native, and network operations. - **Continuous Optimisation**: Maintaining agility and resilience through iterative improvements. --- ## Call to Action for CSPs and Industry - Adopt the CNMM as a guide for AI integration. - Invest in AI literacy and skill development. - Implement AI-ready infrastructure and tools. - Foster collaboration and responsible AI governance. - Move toward fully autonomous, self-optimising, and self-healing networks powered by Agentic AI. --- ## Conclusion The integration of Agentic AI into cloud-native telecom operations is a critical step toward next-generation, autonomous networks. By aligning AI adoption levels with CNMM maturity levels, CSPs can systematically evolve their operations, ensuring governance, security, and compliance throughout the process. The document emphasizes the need for a holistic transformation, involving people, processes, and technology, supported by industry collaboration and continuous innovation.