people · May 21, 2026
LangChain details build of LangSmith Engine with subagents and Deep Agents on May 19 2026
Share the canonical public link.
Palash Shah published details on May 19 2026 about building LangSmith Engine as an agent that spots recurring issues in traces and suggests actions. LangSmith Engine uses Deep Agents, a LangSmith sandbox, CLI for compressed trajectories and subagents for screening and investigation phases to handle large trace volumes and update an Issue Board with categories, severity, traces and proposed actions such as evaluators or fixes. It consumes Agent Overview instructions, traces, existing issues and optional codebase then proposes evaluators via code or LLM-as-judge plus regression examples. LangSmith Engine was tested internally at LangChain before public release and turns manual trace review into grouped issues with actionable outputs.