Engineering tasks you can do with Gemini 4 Argon in Google Antigravity

Software engineering is evolving at an unprecedented pace. Between managing legacy systems, localizing platforms for global audiences, and optimizing massive codebases— engineering teams are constantly stretched thin.

With Gemini 4 Argon, developers can hand over complex, multi-turn, and system-level engineering tasks that previously demanded days of tedious manual execution. Argon is effective at managing context, planning the execution, writing the code, running the tests, and fixing its own errors along the way.

Here are 3 ways you can use Argon in @Antigravity to accelerate development cycles, modernize infrastructure, and scale operations.

1. Modernizing legacy codebases

Organizations and scientific institutions often have decades of invaluable research trapped in ancient, undocumented codebases. Rewriting these systems to modern standards manually is slow, costly, and highly prone to regression.

We tasked Gemini 4 Argon with mapping and migrating decades of scientific research written in 1980s Fortran. Operating in Antigravity, Argon autonomously ingested the legacy Fortran codebase, mapped its architecture, found 6 legacy bugs, and rebuilt the entire system from scratch in modern, memory-safe Rust. The newly compiled codebase ran 2x faster, completely eliminated memory-safety risks, and provided scientists with a clean and maintainable foundation for future research.

2. Localizing websites and documentation

Keeping developer documentation accurate and localized across multiple regions is a notorious bottleneck for global launches. We shrank weeks of manual documentation translation down to a few hours with Antigravity.

See how a DevRel Engineer at Google localized 100+ pages of developer docs into 17 languages overnight. Instead of waiting for a traditional translation, he used Gemini 4 Argon and /teamwork-preview to automate the entire pipeline for antigravity.google

3. Performance profiling & optimization

Performance engineering is notoriously difficult. Finding slow spots and framework overhead in a production repository often takes days of digging, which means it rarely makes the active roadmap.

To solve this, our engineering team gave Argon a one-paragraph prompt to eliminate framework overhead in google/adk-python (our open-source AI agent toolkit) without breaking a single existing application. Left entirely to its own devices overnight, the agent profiled the runtime, identified bottlenecks, and optimized real production code. The result was a 40% drop in in-memory session prep time and 15 newly identified bugs. Those fixes are now live in the official open-source repository.

Get started

Gemini 4 Argon works longer and harder to solve highly complex scientific, coding, and agentic problems. Start building with Argon in Antigravity, available for Google AI Ultra and paid Pro subscribers.

Read the blog to learn more about what you can do with Argon.