LangChain released a detailed methodology for iteratively improving AI agent harnesses using evals as a hill-climbing signal. The recipe covers sourcing and tagging evals (hand-written, production traces, external datasets), splitting per behavioral category into Optimization and Holdout sets to prevent overfitting, and running autonomous improvement loops. The key insight: agents tend to overfit to specific tasks during autonomous hill-climbing, so holdout sets are critical for ensuring generalizable improvements.
2026-04-08 / DISPATCHArcade.dev Tools Now in LangSmith Fleet: 7,500+ Production-Ready MCP Tools with Per-User AuthLangChain integrated Arcade.dev's library of 7,500+ agent-optimized tools into LangSmith Fleet, delivering the largest single injection of production-ready tools into the MCP ecosystem. Arcade provides per-user, session-scoped authorization with least-privilege enforcement at runtime — each action inherits the permissions of the specific user the agent acts for. The integration includes 60+ production-ready templates for sales, marketing, engineering, and support use cases, addressing the persistent enterprise pain point of connecting autonomous agents to dozens of SaaS apps (Slack, Salesforce, Jira, etc.) securely. 2026-03-19 / AGENTSLangChain + NVIDIA Launch Enterprise Agent Platform: LangGraph + NIM Microservices Deliver 2.6x ThroughputOn March 16, LangChain and NVIDIA announced an enterprise agentic AI platform combining LangGraph, Deep Agents, and LangSmith with NVIDIA NIM microservices, Nemotron models, NeMo Agent Toolkit, and Dynamo inference engine. LangSmith has processed over 15 billion traces and 100 trillion tokens; NIM microservices deliver up to 2.6x higher throughput versus standard deployments across cloud, on-premise, and hybrid environments. NeMo Guardrails integration enforces content safety and policy compliance at the agent layer — the first platform to bundle observability, guardrails, and inference optimization in a single enterprise offering.