The GLIDE library offers a unified open-source solution for reliable evaluation of agentic systems by providing debiased estimates and valid confidence intervals through prediction-powered inference methods. This tool aims to reduce the costs associated with human annotation while maintaining precision in evaluations, making it a significant advancement in the field of AI evaluation.