EXANTE, an international brokerage platform for trading on global financial markets, has built artificial intelligence directly into its software testing process, using it as a working partner for testers rather than a replacement.
The approach, detailed by EXANTE QA engineer Dzianis Talstsiuk, who is responsible for the firm’s desktop trading terminal, web terminal and mobile applications, aims to broaden test coverage and speed up releases in a domain where mistakes carry real financial and regulatory consequences. A price rendered incorrectly in an order book, or an order left stuck in an unresolved state, can lead to client losses.
EXANTE’s stack combines Python and pytest for automated testing, Playwright for its web terminal, and OpenCV, EasyOCR and PyAutoGUI for its desktop application, alongside AI tools from Anthropic and OpenAI. Rather than simply installing an extension, EXANTE has built clear control points between requirements analysis, development, testing and release.
Central to the model is a shift-left approach, bringing QA in as soon as a product owner has fixed requirements, rather than waiting until a task reaches the testing column. The tester runs requirements through a large language model to check completeness, consistency and unhandled edge cases, supplemented with internal context including EXANTE’s own guides, prior feedback and a Model Context Protocol connection to internal documentation. That analysis produces a checklist developers must clear before handing a feature to QA, while test cases are generated from clarified requirements and linked automatically to Jira and test management tooling.
EXANTE has also identified where AI genuinely helps and where it falls short. The tools prove useful for generating test skeletons, refactoring repeated code, spotting debugging patterns and writing supporting utilities. However, EXANTE flags silent failures as the most treacherous risk, where an AI agent confidently returns an incomplete or incorrect answer without signalling any uncertainty. Complex page objects, drag-and-drop automation, negative assertions and OCR template matching also still require manual correction.
Underpinning the workflow is a structured knowledge base split into CLAUDE.md and MEMORY.md index files, organised into five categories covering processes, integrations, formats, methodology and feedback. EXANTE layers this with slash commands, subagents and MCP servers, plus a “council” command that runs multiple independent AI agents over the same requirement, test case or code to surface additional gaps before they reach costlier, later stages.
EXANTE stresses that the pipeline produces a draft, not a finished result, with QA reviewing every generated output. Talstsiuk said the goal is to reinforce human expertise across the team rather than replace it, arguing AI should be treated as a co-worker, not an author.
For more, read the full story here.
Read the daily FinTech news
Copyright © 2026 FinTech Global
