NVIDIA engineers use Codex with GPT-5.5 to automate complex engineering tasks and machine learning research workflows

NVIDIA's engineering and research teams are using Codex (an AI coding agent built on GPT-5.5) running on NVIDIA GB200 and GB300 infrastructure to handle complex tasks including building production systems, writing machine learning experiment scripts, and automating the research loop from hypothesis to execution.

Codex can sustain long, autonomous sessions with multiple compactions while maintaining accuracy and context; it also supports SSH for running large machine learning workloads directly from a laptop, and can translate old codebases (e.g., Python to Rust) making them 20X more efficient.

Senior Software Engineer Dennis Hannusch evolved an internal platform from MVP to production-ready status using Codex, and the team built an internal podcast recording app in hours—work that would have taken weeks through traditional procurement. An AI researcher reports a 10x speed improvement in running machine learning experiments.