Kapoor, S., Stroebl, B., Siegel, Z. S., Nadgir, N., & Narayanan, A. (2024). AI Agents That Matter. arXiv preprint arXiv:2407.01502.
This paper critiques current AI agent benchmarks and evaluation practices, highlighting key shortcomings such as the narrow focus on accuracy, lack of attention to cost, conflation of benchmarking needs for model and downstream developers, inadequate holdout sets, and lack of standardization in evaluation practices. It proposes a framework to address these issues, aiming to optimize both accuracy and cost, avoid overfitting, and enhance reproducibility. The goal is to develop AI agents that are not only accurate on benchmarks but also useful in real-world applications.