Gemini 4 Argon Modernizes 40-Year-Old Scientific Code, Moving Fortran to Rust in 59 Minutes

At a glance

Using Gemini 4 Argon in Antigravity, Google moved the core of a 174,000-line coastal ocean model written in Fortran, a long-standing language for scientific computing, into fully memory-safe Rust, a language designed to prevent memory errors, in 59 minutes. It also released the workflow as an open-source skill anyone can use.

Key points

1. 1

   The AI agent found six bugs that had lingered for decades, halved memory use, and made simulations 1.88 times faster.

2. 2

   Models for hurricane storm surges and weather forecasts still lean heavily on Fortran code from the 1970s and 1980s.

3. 3

   Porting such code by hand takes years, and tiny changes in calculation order often make results drift from the original.

4. 4

   Google says the job needed an agent tool because the whole codebase has to be seen at once, which pasting snippets into a chat window cannot do.

Why it matters

If trusted old scientific code can be moved to modern tools without breaking it, researchers can spend less time fighting outdated build setups, the steps that turn code into a runnable program.