LangSmith changes teased to tackle an AI agent eval bottleneck

One post argues that tweaking prompts and testing a few inputs by hand only gets AI agents so far. It describes evals that grade an agent’s answers or steps, including a process for turning problems found in production into offline test cases. Quoting that post, another poster called evals the biggest blocker in most agent development pipelines and said on October 6, 2026, that changes for everyone in LangSmith were coming in “a few short weeks.”