Svoi.ru reduced test preparation by 70% using AI agents

AI is already helping QA not only write automated tests, but also eliminate the most expensive hidden stage — preparing product verification. The Svoi.ru team demonstrated that AI agents can take on requirement analysis, context gathering, and draft test documentation preparation, reducing time on this stage by approximately 70%.

The problem the team identified is familiar to almost any QA specialist. From the outside, QA work often looks like running scripts and finding bugs, but inside the process, a significant portion of resources is spent long before the first click on the system. You need to read requirements, track down related issues, cross-check documentation versions, understand how business logic actually works, find dependencies between services, and reconstruct the full picture from scattered sources.

If a product evolves rapidly, this preparatory stage begins consuming hours, and sometimes days, especially when information is stored in multiple systems and updated asynchronously. This is where AI proved useful not as a universal test generator, but as a tool for analytical routine work. Instead of attempting to replace the QA specialist entirely, the team focused on a specific narrow area: speeding up information gathering and structuring before testing.

The logic is clear: the faster a specialist gets a complete picture of a feature, the sooner they can move on to risk assessment, scenario selection, and actual system behavior verification. This approach removes one of the main bottlenecks in the QA process — constant switching between requirements, tickets, comments, mockups, and internal agreements that are rarely gathered in one place. According to the case description, AI agents were used as an intermediary layer between the QA specialist and knowledge sources.

They help read input materials, extract key entities, gather task context, and prepare a clear foundation for further work. In this format, the agent is valuable not because it makes final decisions, but because it saves time on searching and organizing data. The QA specialist still bears responsibility for quality, priorities, and final interpretation, but spends less effort on mechanical tasks: copying facts, cross-checking formulations, and drafting initial test documentation.

The 70% effect appears especially significant because it's not about local acceleration of a single operation, but about reducing load on the entire preparatory cycle. When time is spent not on reading dozens of documents, but on meaningful test coverage, the team reaches complex scenario verification faster, finds gaps in requirements sooner, and depends less on manual knowledge transfer between people. Moreover, such agents can also be useful for adjacent roles: analysts, developers, quality managers.

If a single mechanism can gather context and make it readable, not only QA benefits, but the entire change delivery cycle. Equally important is that such a result does not mean automatic replacement of QA specialists. On the contrary, the case shows a more mature AI implementation scenario: not substituting expertise, but amplifying it where humans spend time without adding new value.

Test preparation is a good candidate for such automation because there are many repetitive actions, lots of textual information, and a high risk of losing details during manual compilation of the overall picture. The more complex the product and the more business rules it contains, the more noticeable the benefit from an assistant that quickly consolidates data into a single representation. For the market, this is another signal that the next phase of AI implementation in development is linked not only to code generation.

The most noticeable returns often come from less public, but expensive processes — requirement analysis, artifact preparation, context transfer, and reduction of operational routine. If such practices take hold, the QA role will shift even more toward research and product expertise, while routine preparation will increasingly be covered by specialized AI agents.