A new OpenAI field report details how scientists are leveraging AI coding agents to modernize scientific computing, significantly accelerating software development and discovery, particularly in genomics. Historically, crucial research software, often developed by academics with limited engineering resources, has been prone to fragility and slow maintenance, impeding scientific progress.

AI agents, such as Codex and Claude Code, are beginning to address these challenges by reducing the cost of engineering tasks. This allows researchers to prototype ideas faster, undertake previously impractical projects, and improve long-term software maintenance. The report features eight projects, primarily in life sciences, where agents assisted with tasks ranging from routine updates to major code migrations and GPU redesigns.

Contributors reported substantial acceleration in development and maintenance, enabling small teams to handle complex tasks. A key shift observed is the researcher's role evolving from implementation to verification and orchestration, with humans retaining control over scientific direction and quality. However, challenges remain in establishing clear, long-term stewardship for AI-generated tools, emphasizing the continued need for human judgment in validating AI output and ensuring the durability of scientific software infrastructure.