Section 5: Customising agents

By now you’ve got an agent doing useful work and you know the security trade-offs of letting it run. The next question is how to make that behaviour repeatable — packaging what you’ve learned into a skill the agent can call by name, or splitting a task across specialised sub-agents — rather than re-explaining the same thing from scratch every session. We’ll build both around the same running example: fitting, plotting, and writing up the topa/coral-cover GLM.

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.