Hitting Reset on Automation: Knowing When Custom Scripts Aren’t the Answer

With expanding backlogs and increasing pressure to enhance discovery and accessibility, automation can feel like a godsend. Using generative AI to create Python scripts, I built a suite of small tools that now handle routine tasks like bulk renaming, file movement, folder restructuring, and Excel manipulation shortening tasks from hours to seconds. Encouraged by that success, I attempted something more ambitious: automatically extracting table of contents data from 70 issues of a digitized historic literary publication to enrich metadata at scale.

Initial tests showed promise, but inconsistencies in layout, labeling, and page placement quickly caused the script to fail. After multiple iterations and narrowed goals, the output remained too inconsistent to scale across the collection. We pivoted to an AI-powered metadata assistant within our library system, which produced cleaner results but introduced new workflow constraints and still required manual cleanup. Ultimately, we paused the project.

However, the same script worked exceptionally well on a more standardized publication, demonstrating that automation’s success depends heavily on consistency in source material.

This lightning talk reflects on when automation meaningfully advances access, and when “hitting reset” is the more sustainable choice.

SLIDES