Tagged
Agents, middleware, evaluation and releases across the LangChain stack. Curated and summarized from dozens of sources by AIBriefs.
Jev is a non-generative "System One" model that returns typed answers and probabilities instead of tokens; TypeSafe AI reports up to 200x faster inference and 400x lower cost than comparable LLMs on classification tasks. It is trained with reinforcement learning for calibrated decisions (RLCD).
Open source agentic assistant pulls from 600K+ ClinicalTrials.gov studies, 29M PubMed abstracts, and 12M PubMed Central full-text articles. It runs data analyses in a secure sandbox and delegates tasks to hundreds of sub-agents.
Included Health's Dot uses a federated multi-agent architecture built with Deep Agents, LangGraph, and LangSmith for healthcare navigation with human handoff and clinical oversight. Engineering manager Kartik Darapuneni said the prior structured navigation trees felt "really rigid" to members.
LangChain's guide covers three healthcare and life sciences agent programs: Madrigal Pharmaceuticals' enterprise multi-agent platform, Abridge's clinical documentation agent, and Vizient's GenAI platform for querying siloed hospital data.
The Deep Agent lives in Slack and reads spend, clicks, and conversions across 6 ad platforms plus LangChain's data warehouse every Monday, then proposes changes it applies once approved. LangChain scaled from organic growth to five paid channels in six months before building it.
Credit Genie's AI/ML teams use OpenWiki, LangChain's open-source repo-documentation agent, to auto-generate and update docs from code changes. Nightly runs, automated PRs, auto-merge workflows and daily GitHub Pages portal rebuilds keep docs fresh without manual upkeep.
Connections in Managed Deep Agents v0.7.0+ stores credentials in LangSmith workspace instead of .env, supporting agent-owned secrets and user-owned OAuth. It handles the OAuth round-trip, so no callback route or token store is needed.
LangChain's deepagents now supports context modes "isolated" and "fork", letting subagents inherit a supervisor's full conversation instead of starting from a blank window. The blog says forking is faster and cheaper than isolated subagents because it reuses the supervisor's conversation via prompt caching.
TradingAgents is an open-source framework using multiple LLM agents to simulate a trading firm for financial analysis and decision-making. It integrates real-time data and supports various LLMs.
MCP support now lives in langchain.mcp, built on FastMCP for the 2026-07-28 spec, with elicitation handled as a LangGraph interrupt and tool lists cached. MCP tool calls from ChatGPT users are up 98x across 2026, having more than doubled in August alone.
Schneider Electric runs an internal AI Hub of 350 people supporting more than 60 agents across critical infrastructure. monday.com rebuilt its Sidekick assistant into layered subagents and sandboxes after finding more tools made it worse; Vodafone built Insight Engine and Enigma on LangGraph.
Nevermined's agentic payment infrastructure lets LangChain agents delegate a credit card with spending limits to buy web search credits, top up tokens, and purchase access to new providers mid-task without human intervention. Every payment is traced on LangSmith.