Several studies have examined how ML practitioners at Big Tech companies or in academia do their work, including the workflows they use and the problems they face. These studies often inform the design and development of new tools, frameworks, and platforms. However, fewer studies have considered the practice of ML practitioners at smaller organizations. In this paper, Aspen Hopkins and Serena Booth extract insights from 17 interviews with technologists at startups and smaller public companies. They find that companies facing resource constraints often encounter significant challenges in developing and testing fair and robust ML models. These challenges include accurately predicting performance and cost, evaluating and iterating on models and datasets, mitigating bias, and interpreting privacy regulations, among others.