LangChain

LangChain is hiring a principal/lead engineer in Cambridge, MA to help shape the technical direction of the LangSmith agent engineering platform. Please reach out if you're interested or know someone exceptional who might be! https://lnkd.in/e4GuaGVd

The latest episode of the Max Agency podcast has dropped. Check out Harrison Chase’s conversation with Cogent Co-Founder & CTO Geng Sng ⤵️

Join us for a technical fireside chat in NYC with Harrison Chase and Anish Agarwal (CEO of Traversal) on June 2nd. RSVP: https://lnkd.in/gg8_RcAU Harrison and Anish will dive into how Traversal builds, ships and improves their agents from architecture decisions to how they monitor and test their agents in production. Enjoy drinks, food from a live chef 👨🍳 , and engaging discussions! Spots are limited ✨ 

If your AI agent has rules it can't break (and most do), you need to work with the experts who own them. But do you speak legalese? I don't. Do your lawyers think in eval datasets and judge prompts? Didn't think so. So how do you align with people who don't speak your language? How do you make evals the central alignment surface for the build? At LangChain Interrupt 2026 I walked through what worked for us at Chime: a shared taxonomy of risks, expert-authored definitions that directly translate into datasets and judges, and a flywheel where every annotation improves the system. If you are working on similar problems, please reach out, I'd love to chat. https://lnkd.in/dAWQfYVJ

🎓 The hardest truth about building agents? You don’t know what they’ll do until they’re in production. https://lnkd.in/gsRtiBnD Agents aren’t built like traditional software. You can’t test them against a known set of inputs and reliably predict outcomes. When your users can say anything and you’re working with non-deterministic models, it’s impossible to test your way to full coverage before launch. To build great agents, you need to understand how they behave in production by analyzing conversations, responses, and execution steps. We built LangSmith to solve the biggest challenges when monitoring and improving agents in production, and we recently released a course that will help you master our platform. Register at the link above to enroll in LangChain Academy for free, and get ready to start shipping amazing agents faster.

Databox Software Engineer Luka Lovenjak wrote an inside look on how his team uses LangSmith to evaluate their multi-turn analyst agent Genie. Read on to learn: ✅ How to design pairwise experiments that isolate what the agent did, how it communicated, and whether the user got what they needed ✅ How to use a simulated user with a turn budget to measure both effectiveness and autonomy ✅ Best practices for building and improving evals https://lnkd.in/ggzeYgxQ

Introducing the sandbox Auth Proxy: A way to control the boundary between agent-generated behavior and the rest of the world. https://lnkd.in/ghycfU2k Agents need credentials to do real work. But handing API keys to a runtime that executes arbitrary code can be a real risk. With the sandbox Auth Proxy, sandboxed agents can now call external APIs without ever possessing the credentials. Now, you can give agents access to the systems they need, while keeping credentials and network policy under platform control. Take a deep dive at the link above.

Building useful deep agents is the easy part. Running them in production is where teams stall. Every team ends up wiring the same primitives: a durable harness, long-term memory, sandboxed code execution, auth, observability. Over and over. ✅ A managed harness built on the open-source deepagents — durable execution, HITL, model-agnostic, one-line deploy ✅ Context Hub: A managed memory and context layer wired into your agent by default, with prompts, skills, and subagent definitions versioned and saved ✅ Sandboxes: A default per-thread sandboxes with persistent filesystem, safe code execution, shell access, and built-in auth proxy Write your agent. Set a config file. Deploy. https://lnkd.in/epHbwbqc?

Hunter Lovell from the LangChain OSS team with an excellent explanation of interpreters.

During Interrupt, we announced the General Availability of LangSmith Sandboxes. Here’s what you need to know https://lnkd.in/gTnj3ka3 Agents that use code execution as part of their workflow are exploding in usage. They generate code, install dependencies, run tests, inspect failures, and edit files. These types of agents need a computer-like environment with a filesystem, package manager, shell, and persistent state. They also need isolation, because the code they run may be generated by a model, pulled from an external dependency, or supplied by a user. But the risks of running outside of isolation boundaries can bring potential supply-chain attacks in your runtime, broken kernels, and more. We built LangSmith Sandboxes specifically for these demands.  ✅ Every agent get a computer-like environment they can use without putting infrastructure at risk ✅ Managed through the same LangSmith SDK and API key teams already use ✅ Compatible with Deep Agents, Open SWE, LangSmith Deployment, LangSmith Fleet, and custom code ✅ Production controls teams need around credentials, resource limits, lifecycle, and access ✅ New capabilities for parallel workloads, cost control, and enterprise security You can start using LangSmith Sandboxes today with one line of code with your existing SDK and API key. Visit the link above to learn more about our GA release and see how you can get started.