NVIDIA engineers have adopted Codex built on GPT-5.5 as their default tool for complex engineering work, running it on GB200 and GB300 infrastructure. The system handles extended autonomous sessions, surfacing bugs and program gaps that other models missed. Dennis Hannusch, a senior software engineer, praises Codex for its autonomy and ability to maintain context across multiple compactions, revolutionizing how they approach development tasks.

The coding agents team at NVIDIA has leveraged Codex to transform an internal platform from MVP to production-ready, improving scalability and reliability. They also built a podcast recording app similar to Riverside in just hours, a task that would have taken weeks under normal procurement processes. With computer interaction capabilities, Codex autonomously built and tested the application without human intervention, completely changing what's worth building.

For research teams, Codex has automated the entire ML workflow, from identifying research areas to executing experiments on remote machines. GPT-5.5 demonstrates 10x speed improvements for running experiments and can rewrite Python codebases into Rust with 20X efficiency gains. Shaunak Joshi notes its creative capabilities in tracing evidence across research papers and building knowledge graphs that help visualize concept connections, accelerating research cycles.