Deploying an AI Agent in Tech Support: The MTS Case

MTS Web Services has built an AI agent that automatically triages tech support tickets. It dropped its RAG-based search system, which required engineers to phrase queries correctly.

Why they built RAG, and why it wasn't enough

The second-line SRE support team for MTS Web Services' advertising technology cluster works daily with four data sources: Jira with incident history, Confluence with documentation, system logs and databases. When handling complex incidents, engineers spent a significant amount of time looking for similar cases and analyzing solutions already adopted.

The first answer was a RAG system: data from Jira and Confluence was merged into a single search base that supports natural-language queries. Engineers could ask questions, and the system returned relevant documents and incident records. The effect was noticeable, and the team received positive feedback.

Over time, however, a structural limitation emerged. RAG is a search tool: for it to work, the user must know what to look for. In incident investigation that condition is not met: the cause of the problem is unknown, and it is unclear which data is needed and in what order to request it. As a result, the tool went unused precisely where help is needed most: at the start of an investigation.

The second problem was the inability to measure the real effect. There was positive feedback, but the team could not obtain objective metrics, such as how much time was saved per ticket or how MTTR changed.

How the agent takes over the first stage

"If the system can already search for information and analyze what it finds, why not entrust it with the entire first stage of an incident investigation?" —

Vladimir Drobot, SRE lead at MTS Web Services.

Instead of a passive tool waiting for a query, the team built an agent that runs the investigation itself. The agent receives a ticket as input and autonomously completes the first stage:

• Analyzes the incident description and builds an investigation plan
• Determines on its own which data it needs for the analysis
• Formulates and runs search queries against Jira and Confluence
• Aggregates information from several sources
• Hands the engineer a structured initial analysis with hypotheses about the causes

The difference from RAG is fundamental: the agent does not wait for a human to ask the right question. It decides what to search for based on the ticket's context.

What this means

The MTS Web Services experience illustrates the key difference between an assistive tool and an autonomous agent in a corporate environment. RAG supports an engineer at the stage when they already know where to look. The agent takes over the part of the work that consumed the most time, forming initial hypotheses and collecting data, and frees the costly attention of L2 specialists for real analysis. This shows that agentic architecture is already practical today for internal corporate tools, not only for public products.

Why are companies adopting AI for tech support?

To automate the initial triage of tickets and save engineers' time. MTS switched to an AI agent because its previous RAG system required specialists to know how to phrase search queries correctly.

How does the MTS AI agent analyze tickets?

The agent processes four data sources: incident history in Jira, documentation in Confluence, system logs and the databases of the advertising technology cluster.

What data sources does the AI agent in tech support use?

The MTS AI agent works with four sources: Jira with incident history, Confluence with documentation, system logs and databases.