LangChain

🔎 Announced at Interrupt: AI Search in LangSmith You used to need an evaluator or read hours of conversations to find traces that answer questions such as: 💬 Does the user seemed frustrated?" 💬 Where did the agent fail to complete the task? 💬 Where is the user trying to cancel their subscription? Now, just describe it. LangSmith returns the traces where it happened. https://www.langchain.com/blog/agent-traces-ai-search-langsmith

How do agents remember? The longer an agent runs, the more it takes in (instructions, tool results, files, and drafts), which results in decreased accuracy and reasoning. Read how LangChain's Deep Agents harness addresses this by moving context out of the active window without deleting it ⤵️

Our education team answered the most common questions we've received over the years about agent evals. Check it out ⤵️

At our recent Learning Loop event, @oneill_c from @baseten explained why continual learning keeps breaking, and what to try. Watch the full session ⤵️

Agents 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.

🧵 Behind the scenes of AI Search in LangSmith: TL;DR 1️⃣ SmithDB turns traces into readable trajectories. Structured filters narrow the candidates when you search 2️⃣ A model checks each remaining trajectory against your description and streams back matches

Claude Platform traces can now be sent to LangSmith through OpenTelemetry. Export from Claude Managed Agents, the Messages API, or the Claude SDKs with native OTel support. Traces show up in LangSmith so you can inspect every step and improve agent performance. It’s available now in public beta. Docs: https://docs.langchain.com/langsmith/trace-anthropic

And that's a wrap on Interrupt London!

ICYMI: AI Search in LangSmith!

"I can finally find the data that I want without having to fight my tooling." LangSmith helps @TrustVanta find traces fast, align evaluators to labeled examples, and improve AI quality end to end. ⏯️ Zheng Tao on why "LangSmith gets it."

How @OrangeBusiness evaluate agents in their enterprise AI platform: ✅ RAG quality: golden datasets, retrieval recall, answer faithfulness ✅ Agent behavior: tool selection accuracy, prompt injection robustness ✅ Production traces: latency, cost, failed runs, human feedback All in LangSmith, with regression evals before changes reach production, and LangChain for model neutrality & portability https://www.langchain.com/blog/how-orange-business-built-live-intelligence-on-langchain-and-langsmith?

LangSmith for Startups spotlight: @tryopine The customer intelligence layer for revenue teams Opine traces every production LLM run through LangSmith, and gates every prompt and model change behind LangSmith evals. See what Opine looks like on your hardest open deal at http://tryopine.com. And if you'd rather help build the context customer intelligence layer than buy it, they’re hiring across GTM and engineering!