Charlotte Patrick Research, October 2025. Author: Charlotte Patrick, independent industry analyst. Commissioned by Vitria Technology. The analysis and conclusions are Charlotte Patrick Research’s.
What an agentic architecture actually requires
The note sets out eight layers that all have to function for an agentic system to work: the human layer at the top, then apps, agents, intelligence, knowledge, data, compute and storage, with an operations layer running alongside for lifecycle, security, governance and risk. The point of framing it this way is that agentic systems are usually discussed at the agent layer alone, while the constraints turn out to sit in knowledge and data.
Three stages, and where most operators will pause
Stage one is a copilot interpreting a human instruction and routing it to a tool. Stage two has agents coordinating across hierarchies for more complex tasks. Stage three grants agents genuine autonomy — technically possible, but with a business model and feasibility that remain uncertain. The research suggests many telcos will find a natural resting place between stages two and three, taking the benefits already available while building out the knowledge and data layers that make the next step viable.
Six questions to answer first
The note structures its analysis around six: how to build an agentic architecture successfully, what the financial benefits actually are, how to supply models with context, whether available data can support training and decisioning, what a multi-agent system is worth, and what it takes to build one. The treatment of the financial question is notably unsentimental — it puts a figure on the upside of reaching Level 4 autonomy and then asks whether that justifies the investment.
The comparison with RPA
The most useful section may be the conclusion, which reads agentic systems against the industry’s experience of RPA. Four years after RPA’s starting point, roughly half of enterprises had not progressed past their first ten bots. Bots broke when applications changed, maintenance was expensive, and returns were clear on high-volume repetitive processes but thin where work was variable or cross-functional. The research draws the obvious parallel and asks what would have to be different this time.
What it looks like in practice
Vitria’s own deployments show the pattern the research describes. An internet service provider cut incidents by 65% within 90 days and began detecting 90% of issues before customers reported them. A large US internet service provider eliminated 250,000 unnecessary technician dispatches a year, saving $16 million in operating expense, by correlating customer care issues with infrastructure problems. A Fortune 200 mobile carrier accelerated its 5G rollout by up to three months by building a knowledge-based assurance system from the ground up. (Vitria Technology, PR Newswire, 2 September 2026.)
Vitria’s foreword: from proactive analytics to the agentic journey
This section is Vitria’s own writing and is reproduced in full. Everything above is a summary of the researcher’s work, in our words. Keep that distinction if this page is edited.
The journey toward agentic systems has a deep historical foundation. For decades, companies have worked to use analytics and automation to transform telecom operations, and Vitria has a long history in streaming analytics and operational intelligence.
Vitria laid the groundwork by focusing on turning large volumes of cross-domain data into actionable insight for automated remediation of network problems — self-healing. The VIA AIOps platform reflects a sustained commitment to helping telcos overcome long-standing barriers to service quality and assurance, by applying AI, machine learning and structured knowledge to detect issues and identify root causes across multi-vendor, multi-domain environments.
The transition to agent systems, which moves beyond scripting to enable hierarchies of agents and, in time, fully autonomous ones, is the next logical step toward autonomous networks. Vitria’s record of driving measurable improvements in service availability and resolution time informs a clear view of what this phase involves. Telcos will continue to need clean, usable data, both for model training and increasingly for real-time dynamic automation. They will also need a knowledge plane, which provides the context that improves agent understanding and supports all other machine learning and AI in the network.
Vitria’s AI with Knowledge approach uses a knowledge plane grounded in knowledge graphs to store and apply contextual, topological and diagnostic information. That is particularly effective where accurate insight has to be derived from high-volume data streams across increasingly complex service delivery layers.
By taking on this era of intelligent automation, telcos can turn reactive systems into autonomous networks that anticipate and prevent service disruption — a shift toward a genuinely adaptive and customer-centric future, and a journey Vitria and its peers have been preparing the industry for over decades.
Who should read it
Anyone building an agentic roadmap, or being asked to justify one. It is more useful for sharpening questions than for confirming plans.
Commissioned by Vitria Technology. The research and conclusions are Charlotte Patrick Research’s.
