LangChain introduced new LangSmith features at Interrupt, including AI Search for semantic trace analysis and Managed Deep Agents v0.9 with Slack integration, user-level memory, and scheduling. The update also includes LangSmith Engine v2 for pre-production issue detection, aiming to enhance developer control over agent building, debugging, and deployment.
Agent Tracing
智能体追踪
Recording the complete intermediate steps and state changes from input reception to output generation
Curated Intelligence
LangChain has launched Custom Apps in LangSmith, enabling developers to build, publish, and run custom UIs using natural language prompts. This feature addresses the limitations of default interfaces by allowing users to create tailored views for agent traces, experiments, and feedback data without writing frontend code from scratch.
LangChain has launched LangSmith Engine, a meta-agent designed to improve other AI agents. It ingests agent execution traces to automatically identify recurring failure patterns and suggest specific optimizations. The post details the architectural decisions enabling it to analyze large volumes of trace data.
Arize Phoenix introduces an integration with the Hugging Face Smolagents framework, offering visualized agent tracing and automated evaluation capabilities. The tool records inputs, processing steps, and final outputs to help developers diagnose agent behavior and quantify performance.