When will AI agents begin to design and deploy new AI apps for OSS/BSS?

Artificial intelligence (AI) is rapidly reshaping industries across the globe, and the telecom sector is no exception. Generative AI (Gen-AI), which encompasses AI models that can produce new content such as text, images, and code, and large language models (LLMs), is a subset of AI models focused on understanding and generating human language. It has accelerated this transformation, unlocking unprecedented capabilities for automation, optimization, and innovation. Agentic AI goes further by using AI agents to complete tasks and processes, automating complex tasks at scale. These advancements are particularly impactful in the context of 5G, where the increasing complexity and scale of networks demand more intelligent and autonomous solutions. From dynamically optimizing network performance to personalizing customer experiences, AI is the only way to realize the full potential of 5G and drive the evolution of telecommunications.

At Ericsson, we’re working with foundational Gen-AI agents called “worker agents” to develop AI applications for telecom OSS/BSS. These worker agents are specifically designed to interactively create and deploy new, specialized AI agents tailored for the unique needs of the telecom domain. Think of these worker agents as the architects of a new generation of telecom AI, responsible for bringing into existence agents such as intent based autonomous network agents, network insights agents, business configuration and process work-flow agents, guided selling agents and more.

For years, AI in telecom has focused on prediction, forecasting, anomaly detection, root-cause analysis and optimization which required data engineering, data science, AI engineering in carefully crafting AI models in a domain or an application context. The development, deployment and management of these systems often require significant human expertise and oversight, time, and resources. Gen-AI's ability to learn and adapt and Gen-AI agents ability to plan, reason and automate tasks makes it more versatile and powerful than traditional AI; still, it also requires some human crafting.

What are Gen-AI agents?

Gen-AI agents are autonomous digital systems that, when given an intent or instruction, reason, plan, and break down the task into smaller actionable steps using LLMs. LLMs are usually assisted in reasoning and planning with the help of external knowledge bases and prompt enrichments, ensuring more accurate and context-aware decision-making. These agents further leverage software functions, called “tools”, to execute these steps within other digital systems. Maintaining a “memory” of past interactions — either to provide context within a session or to incorporate mechanisms for self-reflection, allows the agents to evolve because they evaluate past actions, identify areas for improvement, and adapt their strategies over time for more effective current and future actions. Interaction with Gen-AI agent happens through agent interface, which can be a chatbot or another agent.

Figure 1: Gen-AI agent: Building blocks coming together to perform intended action

Gen-AI agents in action: Automate AI application development and deployment

Ericsson is leveraging Gen-AI agents to automate and accelerate the creation and implementation of AI applications for operations and business support systems (OSS/BSS) with Telco Agentic AI Studio.  We envision a set of foundational Gen-AI Agents –"worker agents" as the masters of many more AI agents that fulfil a variety of tasks. These agents are sophisticated LLM-based systems having deep understanding of the telecom domain –whose raison d'être is to develop, test, package and deploy other telecom-domain-application AI agents.

Figure 2: Automate and accelerate AI application development

Download the above graph as PDF

We’ve established the Ericsson Telco Agentic AI Studio as an accelerator to enable agentic AI. Three components fuel the industrialization of worker agents that can build, manage and maintain other AI agents:

1. Intent Layer: The intent layer provides the interface, an intelligent chatbot to interact with worker agents. One can put various requests such as to upload an agentic application, browse through the agent catalog and supporting components or create a new agentic application.
2. Multi-agent system: The multi agent system is the heart of the agentic ecosystem, which provides worker agents which work together to fulfill the intended objective. Key worker agents operate as follows:

   Worker agent	Function
   Upload	Uploading existing agentic applications into the system and use their metadata to build a metadata repository or agent-catalog.
   Create new	Uses the information in knowledge layer to create customized version or variations of existing applications in the agent catalog. It can combine components of an existing application to create a new application or create a new application from scratch using user input along with the knowledge layer.
   Porting	Ports existing application to a different platform or Gen-AI framework.
   Package, testing and deployment	Processes applications to the target platform or Gen-AI framework.
   Observability and explainability	Ensures accuracy of worker agent output and also explains the reasoning behind the outputs.
   Catalog query	Query, view and discover the agent application catalog.
   Operations and maintenance	Manages the ongoing needs of the Telco Agentic AI Studio.

3. Knowledge layer: Knowledge layer comprises of all those components which are required to create new agents or port existing agents. It includes the agent-catalog, a metadata repository of each agent and its components, prompt templates and agent instructions, design rules, ground-truth data and evaluation metrics to ascertain agent accuracy and other internal knowledge and hooks to external knowledge.

Worker agents (or meta-agents) mastermind the design and development of new, specialized AI agents. These specialized AI agents make it easier to sell, deliver and get paid for telecom services. Some of the specialized AI agents are:

• Business intent management agent: This agent automatically manages an interplay between a “customer intent” such as expectation of a requisite experience or price of a product or service with your “business intent” to maximize business objectives such as improved profit margin, guaranteed experience of a premium customer etc.
• Product catalog configuration agent: The agent interactively captures your marketing needs or user demand to create and manage characteristics, complex rules and algorithm for telecom product and service offerings.
• Guided selling agent: The agent personalizes your customers sales interactions, predict customer needs, and recommend the most appropriate products and services to maximize sales and customer satisfaction.
• Intent based autonomous network agent: Given an intent of network behavior, such agents can independently manage and optimize your network operations, respond to events, dynamically adjust network characteristics, re-plan network function allocation that is re-route traffic without human intervention or with human in loop as required.
• Network insights agent: Interactive agents, which allows to extract network insights, identify trends, visualize network performance and customer experience, discover problems and drill-down root causes.

Streamline and fast-track AI innovation and industrialization

Automating AI application development with Telco Agentic AI Studio, encourages application component reusability, optimizes processes and fast-tracks AI innovation. It accelerates customized Gen-AI application development, packaging, testing and deployment. It provides a centralized repository for all applications and application components, ensuring reuse and bringing optimization. Agents-catalog is supported by rich source of internal and external knowledge base.

With an intent-driven approach and agentic AI capabilities, Telco Agentic AI Studio empowers developers to rapidly create tailored Gen-AI solutions. It is built on the Amazon Bedrock platform. The Studio offers a flexible, multi-HCP, cloud-agnostic, hybrid execution environment, rendering it an invaluable tool for transforming Gen-AI concepts into dynamic applications.

Strong governance mechanisms are crucial for success

Ensuring the verifiability and explainability of actions taken by the studio generated agents, especially those based on complex agentic structures and workflows, is crucial for maintaining trust and accountability in critical telecom infrastructure and customer interactions. Robust governance mechanisms take action to ensure that the generated agents align with ethical guidelines, regulatory requirements (like data privacy regulations), and business objectives.

Are you ready to let AI solve real problems?

The convergence of Gen-AI, the rapid evolution of LLMs into sophisticated agents, and the ever-increasing demand for intelligent automation in the telecom sector creates fertile ground for the acceleration of agentic AI. We envision a future where AI not only serves the telecom industry but also actively participates in its continuous innovation and evolution, leading to more efficient, resilient, and customer-centric networks and services, driven by a self-sustaining ecosystem of intelligent agents. The breeding ground for telecom intelligence is emerging, and the first generation of studio created AI agents – managing our networks, understanding our needs, and shaping our services – are now a reality.

Find out more about Ericsson Telco Agentic AI Studio and Ericsson Telco IT AI Apps:

Telecom analytics: transform data into action

Telco IT AI Apps

OSS/BSS evolution for successful 5G monetization

Read more

Maximize OSS/BSS impact with AI and Gen-AI

Driving AI-enhanced business growth Omdia

Artificial intelligence, analytics, and automation

Transform invoice anomaly detection with AI smart encoders