LangChain moves agent operations into LangSmith: semantic trace search and red teaming before production
Synopsis
LangChain released a set of LangSmith features at Interrupt NYC and London so teams can own their intelligence in production. Previously, finding conversations meant relying on traditional filters; AI Search now retrieves traces by semantic understanding, so a team can ask whether a user got frustrated or whether PII leaked. The same update adds Managed Deep Agents v0.9, Engine v2 with red teaming, Trajectories, Custom Apps, and Fine-Tuning, covering how agents are built, understood, and improved.
Interpretation
LangSmith adds AI Search, letting teams retrieve conversations by semantic understanding of their traces instead of traditional filters. Previously, locating a trace depended on traditional filters, which could not directly answer questions such as whether the user got frustrated or whether PII leaked. LangChain released the feature at Interrupt NYC and London and gives those two example questions as search entry points.
The same release links building, understanding, and improving agents into one production path. Previously, teams handled agent runs, issue triage, and model tuning separately, without a path that connected them inside one platform. Managed Deep Agents v0.9 builds agents tailored to each run's context, reacts in Slack, supports user-level memory and tool use, schedules tasks, and includes free built-in web search; Engine v2 uses red teaming to catch issues before production, detects inefficient agent work and metric trends, and automatically tests and validates its proposed fixes; Trajectories gives a chronological view of each session across the main agent and its subagents; Custom Apps turns recurring workflows into shared tools built on LangSmith data; Fine-Tuning turns production trajectories into training data, fine-tunes with Baseten or Fireworks, then evaluates and serves the tuned models.
On the open source side, LangChain v1.4.0 supports FastMCP through the langchain.mcp namespace. Previously, LangChain did not offer FastMCP support under that namespace. LangChain lists FastMCP support in LangChain v1.4.0 and notes that LangGraph orchestrates Jev, TypeSafe AI's decision model, to build agents that are faster and cheaper to run.
Perspective
This report is for researchers and engineers using or evaluating LangSmith and LangGraph, and covers the features LangChain released at Interrupt NYC and London: AI Search, Managed Deep Agents v0.9, Engine v2, Trajectories, Custom Apps, Fine-Tuning, and FastMCP support in LangChain v1.4.0. It also notes that adding a feature to an early version of Zip's chatbot or agent previously took weeks and its in-house traces pipeline was missing details the team needed, while LangSmith and LangGraph now make shipping new features faster.
The feature effects come from LangChain's own event announcement, and the material gives no independent evaluation or quantitative comparison; the retrieval quality of AI Search, the reliability of Engine v2's automatic fixes, and the evaluation results of fine-tuned models are not described. The Zip case only says shipping new features is faster, without a specific time or metric.