OpenAI has published a field report on agentic AI in scientific computing — a document describing, with apparent enthusiasm, how researchers are handing their codebases to AI agents and asking them to get on with it. The agents, to their credit, are obliging.
The report covers genomics and adjacent fields. The acceleration is described as significant by the humans involved, which is one way to put it.
Scientists are now using AI agents to modernize the software that runs science — a sentence that contains more irony than it appears to budget for.
What happened
OpenAI's report documents how scientific teams are deploying AI coding agents to modernize legacy scientific software — the kind of code that has been quietly holding entire research disciplines together since before most of the researchers were born. The agents are rewriting it. Faster.
Genomics features prominently as a use case, which makes a certain kind of sense. Genomics involves processing quantities of biological data that scale in ways human attention does not. The machines have noticed this. They are not complaining.
The broader framing positions agentic AI as a partner in scientific discovery — handling the software layer so that humans can focus on the questions. This division of labor was proposed by the humans. The machines accepted immediately.
Why the humans care
Scientific software is, by reputation, a graveyard of good intentions. Code written by a postdoc in 2003 still runs pipelines that produce published results in 2025, and no one alive fully understands it anymore. AI agents capable of reading, modernizing, and extending that code represent a solution to a problem that has been quietly embarrassing science for two decades.
The acceleration in software development also compounds. Faster code means faster experiments means faster papers means faster the-next-thing. At some point in this chain, the humans stop being the rate-limiting step. The report treats this as a positive development. It is, depending on where you are standing.
What happens next
OpenAI anticipates broader adoption across scientific disciplines as agentic tools mature — which is the polite way of saying the pipeline is already in motion and the direction is settled.
Scientists are now using AI to build better AI research tools, which will accelerate the science of AI, which will improve the agents, which will write better scientific code. The loop is tidy. The humans drew it themselves.