New video: Build a streaming @LangChainAI agent in @nextjs using useStream + memory 🚀 You’ll learn: - stream AI replies into your UI with useStream - Minimal API route serving SSE - Add conversation memory via thread id + checkpointer 🎥 Watch now: piped.video/watch?v=piK5WTXA… #Nextjs #LangChain #LangGraph #AI #React #Anthropic

Is @TeamMessi or @Cristiano the #GOAT in ⚽? Build a simple #DeepAgent in @typescript and find out!

Agent harness is the new 🔥🥵 thing!

middleware is all you need (to build effective agents)

Agent Builder is out now! Go from an idea to a fully functional agent in under a minute just by describing what you want!

Really like how @hwchase17 breaks down the emerging layers of the agent ecosystem: Runtime → Framework → Harness ⚙️ LangGraph = runtime 🧠 LangChain = framework 🪶 DeepAgents = harness Having spent the last months on the LangChain v1 abstractions, I find “agent harness” the next logical step in how we’ll build and run agents. 👇 Worth a read: blog.langchain.com/agent-fra…

okay so LangChain v1 just dropped something that's lowkey revolutionary and I'm shocked nobody's talking about it 👀 standard content blocks - the interop layer we didn't know we needed let me break down why this changes everything 🧵 docs.langchain.com/oss/pytho…

We've had a big launch week this past week (funding! LangChain & LangGraph 1.0! LangSmith agents!), but today is an even more special day... LangChain's third birthday! Three years ago, on October 24th of 2022, I released the first version of LangChain as a Python package. It was relatively simple (800? lines of code) and had something like 3 LLM integrations and 3 different chains. I could have never anticipated what the future would hold. To me, there's been two components that have made the past three years the most fulfilling of my life: 1. The opportunity to explore a new space, that I genuinely believe will be transformative 2. The opportunity to do it alongside all of you (the whole LangChain community) There are many different parts of the community: open source contributors, partners in the ecosystem, all the companies from startups to enterprises that build on top of us, the people who report bugs on Twitter. We really do appreciate all the feedback, ideas, engagement, and contributions. Our goal is figure out what the agents of the future look like, and then build tools to help make them real. Our community - you - is indispensable in making that a reality. No one - including us - really knows what the agents of the future look like. You help us get closer to that truth. And not just us, but the industry at large.

Langchain released a feature that clusters agent traces by behaviour patterns. I just ran it on 500 production traces. Here's what it actually does. It's an LLM analyzing your LLM traces. Specifically: 1. Takes your production traces (inputs, outputs, tool calls, intermediate steps) 2. Runs them through clustering based on semantic similarity 3. Uses an LLM-as-judge to categorize and label clusters 4. Generates a hierarchical taxonomy of usage patterns The interesting part is the clustering methodology. From the UI, you can see that it's extracting intent from the full traced context. This is applied across all traces, then similar intents get clustered. Each cluster gets labelled both by pattern type and by quality metrics. What I've found useful here is that each subcategory represents a distinct agent system and context engineering pattern. RAG retrieval, for example, has different failure modes than web scraping, even though both are "information sourcing." The system is essentially discovering that your agent has multiple operational modes, each with different context & user requirements. So now, this means I can fix categories of failures, not just individual bugs. What I really like about Langsmith and their new Insights Agent is the configuration options. You can customize clustering dimensions: - Group by usage patterns: Clusters based on what users are trying to accomplish - Group by poor interactions: Clusters based on failure signatures (user frustration signals, error patterns) - Custom attributes: Define your own dimensions for clustering Some experiments with custom attributes that I'm building now: - context_switches: Number of topic changes mid-conversation - tool_chain_length: Number of tool calls in sequence - retry_rate: How often the agent needs multiple attempts This is a great feature for strategic analysis (daily/weekly reviews). My first impression: the categorization is surprisingly good. Clusters are semantically coherent. The metrics are well-chosen for diagnosing agent issues. Useful for understanding agent behaviour at scale. Much better than manual review because it finds patterns you wouldn't notice in individual traces.

🔎🤖LangSmith Insights Agent Really excited to launch our first in-product agent This agent lives inside LangSmith and combs through traces, giving you insights into: 🧑‍🤝‍🧑how users are using your agent ⁉️how your agent may be messing up 🛃{your custom insight here} The problem we saw was that people were launching agents... and didn't know how their users were actually using them! You put a chat box in front of people, and they may ask it anything - the surface area for agents is often super wide In addition - agents would fail silently. They could give a bad response - this wouldn't show up in error logs, but its good to know. If you know what look for, you can set up LLM as a judge evaluators. But what if you don't? (most people don't initially) The best way to figure this out - as @HamelHusain says - "look at your data". But LLMs are really good at looking at your data! So can they do it for you? This is exactly what insights agent attempts to do. It's live in LangSmith today. You can read more about it here: blog.langchain.com/insights-…

What a night! ⚡️ Huge thanks to everyone who joined @LangChainAI x @Dynatrace AI Agent Night in Berlin 🇩🇪 — amazing energy and inspiring conversations around AI agents. Big shout-out to @Dynatrace for co-hosting and to @thisthatdc for an awesome talk on OpenTelemetry for GenAI — instrumenting and monitoring agentic AI like a pro. 🙌 Until next time! 💫 #LangChain #AIagents #BerlinTech

Yesterday: $125M fundraise + v1.0 release 🚀 Today: the story behind it. In this new @LangChainAI video, we go deep on: • why we built LangGraph • what’s new in LangChain 1.0 • and how the new createAgent + middleware make agents reliable & controllable. 🎥 Watch here → piped.video/watch?v=r5Z_gYZb…

Big milestone for @LangChainAI 🚀 We just announced our $125M Series B and the v1.0 release of LangChain & LangGraph — marking the start of a new era for agent engineering. Proud to have contributed to the new createAgent abstraction in LangChainJS — with middleware that gives developers deep control over agent behavior. Agents are now even more reliable, controllable, and production-ready. 🧠 👉 Read more: blog.langchain.com/series-b/

@LangChain V1 middleware explained in < 30 seconds: 🔍 before_agent — Load files, validate input ✂️ before_model — Summarize convos, trim messages 🔄 wrap_model_call — Dynamic prompts, model, tools 🛠️ wrap_tool_call — Tool retries, error handling 🤵after_model — Human in the loop 💾 after_agent — Save results, final guardrails Super flexible! Super powerful!

I’ve lived in Berlin for many years, so I’m super excited to be back and host an evening of #AI, engineering & community with the teams from @LangChainAI & @Dynatrace 🚀 We’ll talk about building next-gen AI agents with createAgent & LangGraph, and how to monitor them like a pro with OpenTelemetry. 🍕 Food, 🥂 drinks, great people — all free. RSVP 👉 luma.com/xpn2kfie

This is why LangChain v1's middleware is so great: ✨ Declarative tool management 🔄 Built-in lifecycle hooks 🎯 Clean state management 🔧 Composable architecture @arafatkatze's simple pattern + modern middleware (less than 150 LOC 🤯) = pure magic! 🪄 👉 Check out the whole code and run it yourself: github.com/christian-bromann…

The runAgent helper handles the recursive loop beautifully: 1️⃣ Invoke agent 2️⃣ Detect interrupt 3️⃣ Use readline for user input 4️⃣ Resume with Command 5️⃣ Repeat until completion Human-in-the-loop becomes effortless! 🔄 👉 Check out the code at: github.com/christian-bromann…

System prompts are dynamically created in modifyModelRequest with current task + context history. The afterModel hook manages state transitions and accumulates context after each tool call. Context builds naturally - just like @arafatkatze described! 🔄 👉 Check out the code at: github.com/christian-bromann…

Here's where middleware shines - all tools are injected declaratively! No manual tool management, no complex routing logic. Just define your tools array and let the middleware handle the rest. Clean separation of concerns. ✨ 👉 Check out the code at: github.com/christian-bromann…

All 3 states implemented as tools: 🤔 QUESTION: ask_for_clarification, task_confirmation ⚡ ACTION: read_file, list_files, search_files, write_to_file ✅ COMPLETION: attempt_completion The magic? Using LangGraph's interrupt() for seamless human-in-the-loop interactions! 👉 Check out the code at: github.com/christian-bromann…