Section 6: Agentic engineering for data analysis

Software engineers have spent decades building practices to catch mistakes before they ship. This section adapts three of those practices — verification principles, red-green test-driven development, and simulated data as a sanity check — for scientific data analysis done with AI assistance.

Connolly, R. M., D. V. Fairclough, E. L. Jinks, E. M. Ditria, G. Jackson, S. Lopez-Marcano, A. D. Olds, and K. I. Jinks. 2021. “Improved Accuracy for Automated Counting of a Fish in Baited Underwater Videos for Stock Assessment.” Frontiers in Marine Science 8: 658135. https://doi.org/10.3389/fmars.2021.658135.