In the AI Era, Coding Means Thinking Like a Systems Architect

Summary

In this Nature Careers column, Jacques Cornwell argues that scientists still need to learn coding even though large language models and autonomous AI agents can generate complex analysis pipelines in seconds. The key skill is shifting from memorizing syntax to understanding available methods, designing robust workflows and critically reviewing automated output. Cornwell recommends thinking like a systems architect, asking AI tools to produce small, single-purpose functions instead of fragile monolithic scripts. He also advocates standardized project structures, such as separating read-only raw data, analysis code and outputs, to reduce the risk of accidental data corruption. Because public cloud models can expose sensitive inputs, he suggests using fictional data that preserves real tables' structure when developing and debugging code, then running the verified scripts locally on protected data. The column presents AI as an assistant that changes the scientist's role, rather than eliminating the need for programming judgment.