In this series of blogs, I have explored the opportunities and challenges associated with agentic AI. The risks, outlined in the previous blog, are essentially rooted in the very essence of agentic AI itself: based on large-language models, it is probability based, determining the next action using odds calculated from the data they have consumed. Clearly the quality, relevance and reliability of that data is fundamental. Agentic AI is also autonomous and progressive; one action leads to another. This can, in the words of McKinsey partner Duarte Begonha, mean that “sometimes agents can get excited and start doing stuff they're not supposed to do." The cascade effects of compounding errors could be cataclysmic if unchecked. As Nokia recently noted in a white paper: “Agentic AI relies on human training and reasoning and can have unwanted outcomes. This means proper guardrails with human supervision of agents should be considered to reduce or eliminate these unwanted outcomes.” So, the question is, what should telcos do today to mitigate these challenges so that they can capitalise on the benefits of this exciting set of technologies at scale?
Power is nothing without control
The solution I discuss here is to marry the autonomous power of agentic AI with the control, context, and trust provided by deterministic data and well-defined ontologies. For agentic AI to move from pilot to enterprise-grade execution, it must be anchored in robust, traceable foundations. Teradata’s rich heritage and expertise in this area, plus our deep engagement with telcos deploying AI has provided an informed perspective on safely and successfully deploying agentic AI at scale.
Two new acronyms are important for any telco looking to deploy AI agents. They are MCP and RAG. Model Context Protocol (MCP) helps provide the essential context to keep agents aligned with the reality of telco operations. MCP supports seamless AI-to-AI and AI-to-tool communication by using existing APIs to connect agents’ reasoning engines to trusted deterministic models, structured data sources, and operational tools upon which the telco relies. Telcos can be confident that agents’ decisions are based on reliable, real-time, and relevant data and executed through established (and permitted) channels.
Retrieval-Augmented Generation (RAG) builds on this to allow agents to dynamically retrieve structured knowledge from vector stores and ontologies. Creating structured linkages between the agent’s autonomous decisions and the governance and deep domain expertise held within ontology-based vector databases provides the confidence that decisions will align with real business goals, critical governance, and contractual obligations.
A prime example is the Teradata Communications Data Model (CDM). This provides a comprehensive, semantic framework for all your data, which autonomous agents can use as a “map.” Using the CDM, AI agents benefit from a common and agreed upon understanding of business entities, relationships, practices, and processes and so interpret, reason, and make data-driven decisions that are also aligned with telco best practices and governance.
Layered decision making
Both these approaches create layered or hybrid architectures that combine agentic AI with more familiar (and trusted) deterministic rules-based systems. Without a deep understanding of enterprise data, user behaviour, and domain-specific nuances, even the most advanced AI agents will fall short. Teradata’s MCP Server implements MCP and RAG approaches that interface with existing enterprise data platforms to support AI agents that can reason, remember, and act with precision. MCP and RAG connect agentic AI to trusted enterprise data, tools, and ontologies. This gives both business owners and data scientists confidence in agent decisions and a clear audit trail explaining how they were made.
For the most sensitive and confidential data, telcos may also want to consider running agentic AI on-premises only. The dangers of data leakage or exposure through AI agents may make keeping data (for example individual customer profiles or network logs) in private rather than public clouds. The power to bring AI agents to where this data resides can also mitigate any potential latency issues.
Humans in the loop
Finally, it is important to implement effective guardrails that ensure humans remain in the loop. As a regulated industry, telecommunications has a high bar for ethical and legal practices. Thus, rigorous testing, human oversight, and transparent decision-making are essential to maintain compliance and customer trust. As Deloitte recently commented, while there are current rules to address general AI safety, bias, privacy, and explainability, there is little guidance for agentic AI systems. Telcos must leverage existing internal governance models and policies while also building new safeguards for human-AI collaboration. A good starting point is to establish exactly what areas AI agents are allowed to manage and put in place stringent boundaries that define what decisions can be made. Differentiated validation layers control how far agentic AI decisions can progress without human intervention, freeing agents to handle large-scale, low-risk work while preventing them from “going wild” in higher-stakes situations.
This is just the first step
It is clear that for agentic AI to deliver significant benefits at scale it cannot be managed as a standalone side project. Instead, it must be deeply integrated into the data fabric of the organisation and share the same principles, governance, and guardrails that apply to all data. Working with Teradata, already a partner of choice for telcos around the world, builds the trusted, scalable, and explainable data infrastructure that agentic AI systems need. The Teradata Vantage platform combined with MCP server and RAG capabilities empowers enterprises to build intelligent agents that unify, analyse, and act on structured and unstructured data at scale.
Today’s applications for agentic AI, from fraud detection and customer intelligence to autonomous network operations, are just the foothills of what could be accomplished. My final blog will look into the future and outline the potential of an agentic AI-driven telco, Stay tuned!