How We Benchmark Deep Agents

We revamped how we benchmark Deep Agents. Here’s the eval setup we run in Harbor across coding, conversation, and retrieval, and how we use it to ship changes.