Why EXANTE rejected AI to migrate its critical tests

Breaking a monolith into microservices is often treated as a simple engineering exercise. For broker EXANTE, a single migration became a full audit of its testing practices. It uncovered false positives, missing coverage and outdated documentation, and it cut regression time from an hour to two minutes.

Grigorii Sedelnikov, automation QA engineer at EXANTE, explained how his team moved the firm’s order risk assessment service, known as Pre-margin, out of its Core monolith and migrated about 200 automated tests along the way.

Core works as a vast calculation engine. It serves hundreds of thousands of clients across more than three million instruments, including shares, bonds, futures and options. Pre-margin checks whether a client has enough funds to open a position, and it runs whenever a trader hovers over the Buy or Sell button. Because these requests far outnumber actual orders, the service needed to scale on its own rather than alongside the entire monolith.

Testing inside Core had become a bottleneck. The framework holds more than 3,500 Python tests split across 16 nightly jobs, and even a small Pre-margin change set off an hour-long regression. Engineers then had to investigate failures in unrelated functionality before any release could proceed. Duty engineers spent four hours a day analysing those failures.

The team first used Claude to catalogue every test that touched the pre-order evaluation endpoints, along with each test’s preconditions and assertions. The job took a couple of hours. Architecture planning stayed largely in human hands, with AI used only to stress-test ideas.

The migration itself was done by hand. In an early trial on one of 19 test files, Claude repeatedly missed account and instrument settings, which is a serious gap in a broker’s environment where small errors are costly. The team also found that writing and validating prompts would cost more than doing the work manually at this scale. The firm’s product manager calls AI “imitation intelligence”.

The manual approach paid off. Engineers found missing setup helpers, oversized test files that blocked parallel runs, untested scenarios and inaccurate documentation. They also found several tests that effectively checked None == None and passed every time without verifying anything. Removing redundant setup code alone cut one run from nine minutes to two and a half.

Today a Pre-margin regression takes two minutes, a 30-fold improvement. Incidents are easier to isolate, duty engineers carry a lighter load, and the team has a clear view of its technical debt.

EXANTE now treats the manual migration as groundwork for future automation. Having mapped every step and pitfall, the team can formalise the process and bring in AI when the next services leave the monolith.

Learn more here.

Read the daily FinTech news

Copyright © 2026 FinTech Global

Enjoying the stories?

Subscribe to our daily FinTech newsletter and get the latest industry news & research

Investors

The following investor(s) were tagged in this article.