The author discusses the lessons learned from building a pipeline for data subject access requests, emphasizing the importance of cost and latency in model-placement decisions. This experience provided deeper insights into designing AI systems compared to other projects.