I. Introduction
Despite the increasing complexity of telecommunications networks, operational processes and workflows remain heavily dependent on manual processes (1). Engineers still spend significant time analyzing alarms, performing diagnostics, reconciling inventories, validating designs, and handling repetitive customer-impact events. This creates a clear gap in efficiency. The problem runs deeper than tooling or talent: as GSMA’s Open Telco AI initiative highlighted at MWC 2026, AI does not yet speak telco — general-purpose models are probabilistic by nature, while network operations demand deterministic, governed, and explainable execution. The industry is still missing a safe bridge between powerful AI and production-grade telecom automation. Tupl bridges the gap by combining explainable AI, multi-source correlation, and expert-driven automation across NOC, customer care, and field engineering—turning expert knowledge into reusable, self-improving pipelines.
II. TuplOS: The governed workspace across network and customer operations
TuplOS is built so your operational knowledge stays with you and grows—one workspace for your network and customer operations. As a no-code platform, it integrates multi-source data (alarms, counters, probes, CRM inputs, configurations, field logs) into a unified operational model with an agentic AI framework, replacing the patchwork of disconnected tools that fragments engineer effort.
Instead of replacing experts with black-box models, Tupl operationalizes their reasoning into explainable workflows. As engineers interact with these workflows, correcting or refining the logic, TuplOS improves continuously.
Designed to lower the AI entry barrier to domain experts, it combines rules, ML, and Gen AI across full operational processes, not isolated tasks—a safe, scalable path toward closed-loop and zero-touch autonomous networks.
Predictive AI models are responsible for streamlining complex processes and automated decision-making, while Generative AI focuses on enriching the user experience. Agentic AI plays a specific, complementary role that will be detailed in later sections.
Two elements are foundational to this mixed approach:
- Turning expert knowledge into digital insights that support better decisions
- Automating manual processes through streamlined workflows and action execution.
This mixed approach reflects a fundamental architectural choice: TuplOS acts as the governed boundary between probabilistic AI and deterministic execution. Generative and agentic AI models are powerful reasoning engines, but they are inherently probabilistic, meaning their outputs cannot be guaranteed to be accurate, repeatable, or safe in every instance. Telecom networks, by contrast, require deterministic, auditable actions: a misconfigured cell or an incorrect provisioning change can affect thousands of subscribers. TuplOS resolves this tension by positioning AI models as creators and accelerators of automation logic, while ensuring that every network actuation is ultimately executed by validated, rules-based pipelines. The result is a platform where AI drives speed and scale, but governance drives safety and trust.
III. Supporting Key Use Cases Across the Full CTIO Chain
Tupl delivers automation across the entire CTIO landscape by transforming expert-driven workflows into repeatable, scalable processes. Each domain benefits from consistent, explainable decision-making powered by TuplOS , which integrates rules, classical machine learning, generative AI and agentic AI in a complementary manner, delivering measurable improvements in performance, customer experience, and operational efficiency.
In Network Operations, Tupl automates high-volume workflows such as NOC triage, incident diagnostics, fault management, configuration and change validation, or inventory reconciliation. By correlating multi-source data and applying expert logic at a massive scale, the platform effectively improves key performance indicators such as mean time to resolution (MTTR), mean time to acknowledge (MTTA), or Network Availability and Change Success Rates.
In energy management, Tupl dynamically handles vendor power-saving features at per-cell granularity, continuously adapting to live traffic and quality thresholds, operating in closed loop and achieving up to 2x more energy savings than default settings or blanket approaches without impacting user experience.
In Service Assurance, Tupl unifies quality of service (QoS), quality of experience (QoE), and customer-impact intelligence into a single automation layer capable of detecting degradations, correlating service events, and assessing customer impact in real time. This changes service assurance from reactive firefighting to proactive experience management, positively impacting key indicators such as Net Promoter Score.
In Customer Operations, Tupl closes the loop between network intelligence and customer-facing processes by automating service activation, complaint handling, and issue resolution. Automated provisioning improves activation success rates, activation times, and SLA adherence, while AI-driven complaint triage and correlation with network events improve FCR, reduce response times, increase NPS, and accelerate complaint closure.
The bridge between network and customer operations is set through Tupl’s Customer Experience Index (CEI), encompassing RAN performance and user-plane data, together with operational context, and customer perception signals. Calculated at per-subscriber and geographic granularity, the CEI enables operators to identify where degraded experience concentrates, prioritize root cause analysis by customer impact, and feed next-best-action recommendations directly into operational workflows including complaint handling, network operations, and planning.
IV. The Agentic AI layer: Accelerating the Path to Turn Expertise into Autonomous Workflows
To understand why TuplOS’s architecture matters, it is worth naming a tension that the telecom industry is only beginning to confront openly: general-purpose AI is inherently probabilistic, while telecom infrastructure demands deterministic, governed, and explainable actions. Large language models and agentic AI frameworks excel at reasoning, correlation, and intent interpretation — but they cannot, on their own, safely actuate changes in a live network where SLA obligations, emergency services, and millions of subscribers are at stake. This is the “AI gap” that GSMA’s Open Telco AI initiative (launched at MWC 2026) formally identified: AI does not yet speak telco, and general-purpose models frequently fail to meet the required levels of accuracy, safety, or operational efficiency.
TuplOS is built around a clear architectural principle to resolve this tension: LLMs and agentic AI create and accelerate automation, but they never become the automation itself. All network actuations are executed by deterministic, rules-based pipelines and validated ML models that sit on the other side of a governed boundary. The probabilistic layer handles reasoning and creation; the deterministic layer handles execution. This distinction is what makes Tupl’s approach telco-grade rather than merely experimental.
Tupl’s approach is to capture expert reasoning, not just patterns in data.
The Agentic Framework Layer empowers domain experts to capture and operationalize their knowledge conversationally, reduces development cycles, and ensures consistent governance across agents and applications.
AI Agents actively participate in the development lifecycle, capturing expert intent and transforming it into working automations. This translates into key benefits such as faster operations (decisions and fixes that once took hours can now be executed in minutes), Smarter Knowledge Use (human expertise becomes a reusable asset), or consistency and Compliance, since Agentic pipelines eliminate human variability, ensuring every action aligns with engineering standards, audit requirements, and SLA commitments.
The Tupl agentic framework is based on a paradigm shift: from tools to versatile workflows based on 4 key elements:
TuplOS Agents
Specialized agents capable of interacting with core TuplOS components (including Tupl Streams, the UI Framework, and the ML Toolkit) to build or modify existing use cases.
Agent Builder
A dedicated TuplOS component that enables the creation of new agents and MCP servers, simplifying the expansion of the automation ecosystem.
Multi-Agent Capability
TuplOS provides a federated multi-agent capability in which agents can call other agents for specialized task execution and coordinated workflows.
The Knowledge Hive
A continuous learning environment that evolves and improves based on insights captured from user interactions, agent behavior, and operational data.
TuplOS Agent builder is the key enabler for domain experts to create agents with prompts, tool orchestration (MCP), and model management, all fully integrated inside Tupl’s platform.
What are the key features of the TuplOS Agent Builder
- Agent Factory – Creates and configures intelligent agents using predefined templates and settings.
- LLM Handler– Manages interactions with Large Language Models, optimizing prompts and responses for accuracy and efficiency.
- Hallucination Detector – Identifies and mitigates inaccurate or fabricated model outputs to ensure data fidelity and safe automation.
- Tool Manager – Oversees all available tools, their configurations, and their accessibility for agents within the TuplOS environment.
- Tool Executor – Executes specific tool functions requested by an agent in real time, enabling dynamic workflow execution.
- Prompt Builder – Generates structured and context-aware prompts tailored to the agent’s task and operational environment.
- Message Manager – Handles the flow of messages between agents, users, and system components, ensuring seamless communication.
- Lessons Extractor – Captures insights and learnings from past interactions to enhance the performance and evolution of agents.
- Context Retriever (RAG) – Retrieves relevant external or internal data sources to ground agent responses using Retrieval-Augmented Generation.
- Summarization Module – Condenses previous conversations or interactions into concise, meaningful summaries that improve context retention and workflow continuity.
V. The Tupl difference:
Tupl is the proven AI automation system for telcos, governed so your expert knowledge stays with you and continuously grows. It provides one workspace for your network and customer operations, delivering end-to-end hyperautomation rather than isolated use cases. Tupl enables full integration of different data sources with a white box approach with Explainable agentic AI.
With Industry alignment to TM Forum ODA and participation in the Ericsson EIAP ecosystem, we ensure an open platform that is easy to integrate with any existing process.
With several solutions already deployed in Tier 1 telco operators in North America, Europe and Japan, these are some proven results TuplOS is already driving:
A Tier-1 North American wireless operator leveraged TuplOS via AI Care Reactive to transform their customer-complaint resolution process: by automating root-cause analysis and handling of technical network issues, Tupl resolved over 80% of escalated customer complaints automatically, nationwide across 70 million subscribers and 400,000 sites. The platform reduced resolution time by a factor of 100× compared to the previous manual process, delivered 100% consistent root-cause detection, cut “No Trouble Found” tickets by 4×, and lowered troubleshooting effort by 90%, thus enabling the operator to serve customers faster while avoiding the need to add more than 100 full-time staff.
A Tier-1 U.S. telecom operator used TuplOS with its Network Advisor solution to automate network optimization and resource management at national scale. The deployment reduced engineers’ analysis time by 67%, auto-closed 60% of reported issues (noise filtering), and cut resolution time for actionable incidents by 30%. Automated, cell-level action plans and AI-driven root-cause identification delivered 100% diagnostic consistency and 90% accuracy, enabling scalable optimization across multi-vendor environments without expanding the engineering team.
A major telecom operator leveraged TuplOS with the Power Saving Advisor to significantly boost network energy-saving outcomes across its mobile and fixed infrastructure. By applying data-driven, automated adjustments—such as dynamic site energy management and intelligent load adaptation—the operator achieved up to 30% additional energy savings beyond earlier manual efforts, while preserving full service reliability and maintaining customer experience standards. The solution also reduced manual interventions, streamlined operations, and supported the operator’s sustainability and OPEX-reduction goals.
These case studies exemplify a more traditional phased approach, in which processes are automated in one process at a time to ensure scalability.
What Agentic AI allows is to accelerate this process to work on a whole transformation program across all operations. We are currently working on this approach with a Tier 1 Operator in Europe, for which we are developing a full E2E automation program through a common hyperautomation layer with TuplOS.
Tupl enables operators to evolve their operations incrementally, one process at a time. Each automated workflow becomes a building block that strengthens the operator’s intelligence, consistency, and self-sufficiency. Each new automated process adds to the operator’s collective intelligence, compounding capabilities, and accelerating the journey toward a self-sufficient network.
Conclusion
By merging governed intelligent automation, multi-source correlation, and agentic AI, Tupl helps operators reduce OPEX, improve customer experience, increase consistency, and move step by step toward true network autonomy.
The recipe is simple:
capture human expertise, automate it at scale, and transform networks into intelligent, self-improving systems, one process at a time.
Explore this content with AI:
Table of Contents
Share this post
