Evaluate your AI agents faster and more effectively

Evaluate your AI agents faster and more effectively What’s changed for agent evaluations? Why you should use evaluations How to get started with agent evaluations About the author Try DigitalOcean for free Related Articles Streamline Your Workflow: Announcing Environment Support for DigitalOcean App Platform GPU Observability: Get Deeper Insights into Your Droplets and DOKS Clusters Image and audio models from fal now available on DigitalOcean By Grace Morgan Updated: December 4, 2025 3 min read Evaluating AI agents can be tricky, especially when your tools aren’t built around how you think and work. That’s why we’re excited to announce that we’ve updated our agent evaluations experience in the DigitalOcean Gradient™ AI Platform. These improvements make it faster and easier to evaluate your AI agents, understand results, and debug issues. The original evaluations feature was powerful but presented friction points that made it hard for developers to adopt. This redesign tackles those challenges head-on: Goal-oriented metric grouping: Metrics are now organized into intuitive, goal-oriented groups such as Safety & Security, Correctness, and RAG Performance. The Safety & Security group is preselected to help developers get started quickly and confidently. Example datasets : A list of example data sets are now available for common evaluations. This allows developers to create their own datasets quickly and efficiently. Clear, persistent error messaging : Upload errors are now clear, persistent, and specific, with messages like “Validation Error: ‘query’ column is missing”. Developers can easily understand and fix issues, reducing friction in the testing process. Interpretable results with trace integration : Results are organized by the same metric groups used in setup, with tooltips to explain each metric and its scoring. Deep integration with observability tools allows developers to jump directly from a low score to the full trace for fast debugging and improvement.