On July 28, 2026, OpenAI published a field report on coding agents in scientific computing, including software modernisation for genomics and other data-intensive fields. The interesting part is not generating a new app from scratch, but working inside existing, highly specialised codebases.
Legacy scientific software can contain years of dependencies, custom data formats and strict reproducibility requirements. An AI agent may accelerate repository analysis, migrations, refactoring and test creation, yet domain experts and experienced engineers still need to verify the output.
The same lesson applies to commercial software. The most valuable capability is often not “build an app from one sentence,” but safely improving a system that already runs: understanding constraints, preserving integrations, protecting data and reducing technical debt incrementally.
Long-term stewardship remains essential. Faster code generation does not replace documentation, component ownership, monitoring or an upgrade strategy. The easier it becomes to produce changes, the more important disciplined review and maintenance become.