Agent evals run through the same graph. Every eval, from single-reply fixtures to simulated multi-turn conversations with LLM patient personas, runs against the real agent, so what they score is exactly what a patient would meet.
@langchain
-
LangSmith Traces and Annotations Drive Agent Improvement Experiments
By
–
Everything is measured in LangSmith. 1️⃣ All conversations land as a trace, are annotated by reviewers, and feed the loop. 2️⃣ Datasets and experiments live in the same place, allowing them to run themed dataset model configs and compare results. 3️⃣ Triage and improvement
-
LangGraph Supervisors Route Health Coaching to Specialist Agents
By
–
LangGraph is the agent. 1️⃣ A supervisor model orchestrates handoffs to domain specialist sub-agents. 2️⃣ Each subagent is focused on one coaching area with its own prompt and tools. 3️⃣ The composition model makes adding new specialists cost-effective and keeps safety evals
-
Nu Builds a Health Coach Within Regulated Healthcare Boundaries
By
–
Nu by @getnuman is an AI health coach in regulated healthcare, built on LangGraph + LangSmith. Here's how their team ships fast without crossing the medical device line 🧵
-
Agent Performance Depends on Models Context and Harness Design
By
–
"it's underrated how hard it is to build agents!" great convo w/ @hwchase17 and @HarryStebbings on @twentyminutevc an agent = model + context + harness, and you can improve an agent by tweaking any of the three. a few takeaways: 1. the highest roi thing you can do is start by
-
A Workshop Teaches Building and Evaluating Life Sciences Research Agents
By
–
Join our EMEA Deep Life Sci Workshop: Agent Engineering for Life Sciences for a hands-on look at how to build, trace, evaluate, and improve research agents for complex life sciences workflows. 🌍 Register for the EMEA session: https://
events.langchain.com/register/emea-
deep-life-sci-workshop/
… -
Harness Squared Improves Agent Harnesses While Keeping Models Fixed
By
–
harness² from @GoogleResearch offers techniques for RSI on the harness, with a fixed model they call it "harnessing", but you might already be familiar w/ this process! it's also called "hill climbing" or "continual learning" more in 🧵 !
-
Technical Session Explores Why Continual Learning Fails and What to Try
By
–
At our recent Learning Loop event, @oneill_c from @baseten explained why continual learning keeps breaking, and what to try.
— LangChain (@LangChain) 16 octobre 2026
Watch the full session ⤵️ pic.twitter.com/USzxElBs5TAt our recent Learning Loop event, @oneill_c from @baseten explained why continual learning keeps breaking, and what to try. Watch the full session ⤵️
-
Managed Deep Agents Keeps Credentials Outside Agent Sandboxes
By
–
Agents with sandboxes run packages you didn't write. If one is malicious, any credential in the sandbox is exposed.
— LangChain (@LangChain) 16 octobre 2026
With Managed Deep Agents, the proxy adds your token outside the sandbox, only on calls to services you've allowed. The agent never sees it. https://t.co/pRb5Tc7EtGAgents with sandboxes run packages you didn't write. If one is malicious, any credential in the sandbox is exposed. With Managed Deep Agents, the proxy adds your token outside the sandbox, only on calls to services you've allowed. The agent never sees it.
-
Managed Deep Agents Provides Security Primitives for Production Deployments
By
–
it's not easy to properly secure production agents!
— nathan drezner (@ndrezn) 16 octobre 2026
we've spent a lot of time building great primitives for Managed Deep Agents to make agents you ship secure by default 🔐https://t.co/rg8ZQE5BuG https://t.co/5DUbwZJXBZit's not easy to properly secure production agents! we've spent a lot of time building great primitives for Managed Deep Agents to make agents you ship secure by default 🔐 https://
langch.in/mda