AivexaNewsSearch
AI news for builders and product teamsChecked every hour

October 2026 LangChain Newsletter: Everything we shipped at Interrupt NYC & London

Collected Oct 15, 2026

LangChain used its October 2026 newsletter to round up everything the company shipped at its Interrupt events in New York City and London. The through-line the company draws is helping teams "own their intelligence" — the model, the harness and the context that adapt a general-purpose model to a specific business — with the bulk of the releases landing in LangSmith, its agent engineering platform.

The headline addition to LangSmith is AI Search, which lets teams find conversations based on a semantic understanding of their traces rather than conventional filters. LangChain gives the example queries "did the user get frustrated?" and "did PII leak?" — questions that depend on meaning rather than exact string matches or metadata tags.

Also new is Managed Deep Agents v0.9. LangChain describes it as a way to build dynamic agents personalized to each run's context, with the ability to react in Slack, maintain user-level memory and tool use, schedule tasks, and more. Built-in web search is included at no cost.

LangSmith Engine v2 targets quality control. According to the newsletter it adds Red Teaming so teams can catch issues before they reach production, and it detects harder-to-spot problems such as inefficient agent work and metric trends. Engine v2 also automatically tests and validates the fixes it proposes.

Two additions focus on making agent behavior legible. LangSmith Trajectories offers a chronological view of each session that spans the main agent and its subagents, so a developer can follow the path an agent actually took through trace data. LangSmith Custom Apps lets teams build, publish and run custom interfaces directly inside LangSmith, turning recurring workflows into shared tools built on their LangSmith data.

Rounding out the LangSmith set is Fine-Tuning, which converts production trajectories into training data, fine-tunes models with Baseten or Fireworks, and then evaluates and serves the tuned models.

On the open-source side, LangChain v1.4.0 now supports FastMCP through the langchain.mcp namespace. FastMCP is a widely used Python toolkit for building servers and clients that speak the Model Context Protocol, the interface standard for exposing tools and data to AI applications; wiring it under langchain.mcp means LangChain developers can reuse that ecosystem instead of hand-rolling MCP plumbing. A separate item points to how LangGraph orchestrates Jev, described as TypeSafe AI's decision model, to build agents that are faster and cheaper to run — a production case study rather than a release.

LangChain also promoted The Agentic Operating Model, a framework for how leading teams combine people, processes and technology to scale production AI. It covers rethinking the agent development lifecycle, continuous evaluation practices, and embedding governance, framed as what it takes to move past experimentation and run agents reliably. Virtual sessions are scheduled for EMEA on Oct 29 at 10:00 GMT / 11:00 CET and North America on Nov 5 at 10:00 AM PT / 1:00 PM ET.

Under its "Speak the Lang" workshop series, Agent Evals 101 runs Wed, Oct 28 at 11:00 AM PT, covering how to measure agent performance by assessing execution trajectory — the sequence of messages and tool calls an agent produces. Agent Memory 101: Context engineering follows Tue, Nov 3 at 10:00 AM PT, a live workshop on how agents remember, using Deep Agents, the open-source agent harness, to show how context windows are kept lean.

LangChain-hosted events include a hackathon with Data Reply in Munich on Oct 27 from 17:00 to 21:00 CET on building with agent harnesses; a LangSmith Roadshow in Chicago on Oct 27 at 10:00 AM CT on building agents with agents; a virtual GTM Engineering Live session Oct 28 at 11:00 AM PT on how LangChain built its design agent; a DMV Agent Builders Meetup in Vienna, Virginia on Oct 29 at 6:00 PM ET; and an open-weight models event in Toronto on Nov 4 at 5:15 PM ET. A community-hosted session in Mexico City on Oct 29 at 5:00 PM CT covers agents in production.

A customer story describes Zip, where adding a feature to an early version of its chatbot or agent previously took weeks and its in-house traces pipeline lacked details the team needed; LangSmith and LangGraph are now said to make shipping new features faster.

Why it matters: For teams already on LangSmith, the interesting cluster is the feedback loop the releases form — Trajectories and AI Search surface what agents actually did, Fine-Tuning turns those production trajectories into training data via Baseten or Fireworks, and Engine v2 red-teams and validates fixes before they ship. My inference is that Custom Apps is the quiet wedge toward consolidation, since recurring evaluation or review workflows can live inside LangSmith rather than in a separate internal tool. The FastMCP support in LangChain v1.4.0 is a smaller but concrete win for anyone standardizing tool integrations on MCP, and it is the one item here available to any developer immediately without a LangSmith plan.

Read at LangChain

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

What we shipped at Interrupt NYC and London: LangSmith AI Search, Engine v2, Trajectories, Custom Apps, Fine-Tuning, and Managed Deep Agents v0.9.