LangSmith has introduced Engine v2, which improves how engineers find and fix agent issues. It can now detect errors, performance problems, and inefficient work. The new version tests fixes against production inputs before deployment, saving time and reducing errors.
Managed Deep Agents now support user-level memory, allowing agents to store context for each user separately. They also support new channels like Slack file transfer and HTTP webhooks, making integrations easier. Built-in web search gives agents access to current information without extra setup.
LangSmith Trajectories offers a new way to review agent sessions. It shows a chronological view of messages, tool calls, and retries. This makes debugging faster and easier for reviewers who did not build the agent. Trajectories help teams improve agent accuracy and performance.
These updates focus on giving teams more control and better tools for building reliable, high-quality agents. They help reduce costs and improve safety by catching issues early and making debugging simpler.
Why it matters
These tools help teams build better agents faster. They reduce the time and effort needed to find and fix problems, saving costs and improving quality.
What to do
Explore Engine v2 features for proactive issue detection. Use Trajectories to review long sessions and improve your agent's performance.
Source: https://www.langchain.com/blog/langsmith-engine-agents-fine-tuning-trajectories
