Modernizing complex legacy code with AI agents.

Mistral AI's legacy-code migration began by making the old and new programs comparable. For a European energy operator's reservoir simulator, the team exported intermediate states from the Fortran program and built C++ tests around those reference values, alongside checks of final outputs. That supplied agents with a concrete way to check numerical agreement while changing the program's structure. Project account.

Mistral says independent translation agents initially preserved the old architecture, while a more structured autonomous workflow later stalled on bugs. The team settled on module-by-module migration with human checkpoints and coding, testing and review agents. The first sprint covered 40,000 lines of a 300,000-line system.

“Build the parity harness before you write migration code—numerical agreement is the cheapest, most convincing proof that a module is done.”

“Get the documentation in order before you lean on the agents, because you can't migrate code nobody can read.”

“It looked like Fortran retyped in C++ syntax rather than modernized code.”

“How to prove the migrated codebase matches the legacy one numerically”