Do AI coding agents save scientists time?

A scientist reflects on whether AI coding agents genuinely save time for researchers. While agents can rapidly produce complete ecological data analyses, they often introduce subtle logical errors requiring deep human review. The author argues that for scientific coding — where understanding the code is essential — autocomplete assistants may be more efficient than full agentic loops. Agents do shine for easily verifiable tasks like custom visualizations or Shiny apps. A referenced study found experienced developers were actually slower when using AI tools despite expecting to be faster. The author concludes that the time-saving benefit is highly context-dependent and that advances need to come in how humans interact with LLMs, not just in model capability.