Legacy modernization projects remain particularly risky, even in 2026. What can AI bring to the table?

In this session, we share a concrete experience report from an AI-powered end-to-end test generation project, designed to secure the refactoring of complex APIs handling €500M in annual revenue. We'll demonstrate how we designed an AI workflow (agents with MCP) to reliably and repeatably regenerate API tests that compare complex JSON payloads exceeding 200 KB.

Moreover, through a short demo, we'll show how the developer role is evolving toward designing prompts that generate prompts that generate code, the value of maintaining strict code quality standards, and how sub-tasks with dedicated contexts can reduce AI costs over time. And because we're in DDD Europe, we'll connect what we learned to Domain-Driven Design practices!

We're sharing an experience report from a 6-month AI innovation project to secure the refactoring of the train booking API for French leader SNCF Connect (unfortunately, we cannot publicly cite the client name), which we conducted for our client.

This is NOT simply development with an AI assistant—it's a fully industrialized workflow that generates and regenerates prompts and code with every API contract change. The API tests replace Postman collections, which were extremely laborious to maintain, and most importantly, enable testing of vastly more cases.

We'll also discuss adaptations for switching between models (e.g., from Claude to Gemini), and how designing prompts/rules/workflows/tasks (making them modular, for example) becomes the new software design in these approaches.

Note: This project was the subject of an experience report presented as a duo with the SNCF Connect Director of Engineering at API Days Paris 2025 in a much less technical version aiming at tech managers.