Are AI systems keeping up with the models?

AI models have significantly improved, making it easier to build impressive systems. However, challenges persist in integrating these models effectively, with issues arising in areas such as context management, retrieval, tool use, evaluation, and monitoring. The core problem often lies in diagnosing why a specific run failed, indicating a gap between advanced model capabilities and the systems designed to utilize them.