2025 in Review – Vicki Boykis (vickiboykis.com)

In a reflective piece on the challenges faced by machine learning engineers, Vicki Boykis shares her experiences since starting a new job as a founding machine learning engineer (MLE). With a focus on building production systems, she likens the process to navigating dark caverns, emphasizing the necessity of grappling with error logs and decision-making amidst constant uncertainties. Boykis highlights the importance of code that must be user-friendly and efficient, revealing the intricate balance between leveraging AI assistance and maintaining context while coding. The significance of this narrative lies in its exploration of the complexities of developing scalable, resilient machine learning systems, particularly the critical role of tokenization in processing data. Boykis underscores that writing clear, effective code requires deep consideration of various possibilities, requiring software engineers to adapt their solutions based on business needs and team dynamics. Ultimately, she advocates for building upon existing libraries and adopting a thoughtful, iterative approach to development, reflecting the ongoing evolution in the AI/ML community focused on practical implementation and user-centric solutions.