Orchard is an open-source framework for the research community to train and evaluate AI agents…

TL;DR - Microsoft Research announced Orchard, an open-source framework for training and evaluating AI agents across multiple task types on shared infrastructure. It matters because fragmented, task-specific agent tooling is a major friction point for reproducible agent research.

• Positioned as a unified research framework covering both training and evaluation of agents, rather than evaluation-only benchmarking.
• Emphasizes infrastructure reuse across task types, reducing per-task engineering complexity for researchers.
• Claims it supports strong performance from smaller models, implying a focus on cost-efficient agents rather than frontier-scale ones.
• Content is a short announcement post (with linked video); no benchmarks, architecture details, or quantitative results are provided.