LangChain Launches Managed Deep Agents for Developers
LangChain has launched Managed Deep Agents to simplify the deployment of production-ready AI agents by bundling runtime infrastructure, sandboxes, and context management into a single service.
LangChain has officially launched Managed Deep Agents, a new developer-focused platform designed to simplify building, running, and deploying production-grade AI agents. This release builds upon the company's original Deep Agents harness launched nearly a year ago. By transitioning to a managed agent model, LangChain aims to address the complex infrastructure challenges developers face when moving agents from local environments to production.
The evolution of agent building has progressed rapidly since LangChain launched in October 2022, followed by ChatGPT in November 2022 and AutoGPT in early 2023. More mature frameworks like LangGraph, Google ADK, and Vercel's AI SDK emerged in 2024 and early 2025, giving developers greater control. By mid-2025, models became capable of running in loops to call tools, powering early agent applications like Manus, Deep Research, and Claude Code. This shift highlighted the need for robust agent harnesses and standardized infrastructure.
Running agents in production requires managing runtimes, streaming events, executing untrusted code in sandboxes, and handling authentication. Managed agent services solve this by bundling the agent harness with infrastructure, utilizing emerging standards like AGENTS.md for instructions and MCP for system integration. LangChain previously explored this via Fleet, a no-code platform for non-developers. Other industry entries include Claude Managed Agents, which introduced concepts like dreaming, and Vercel's Eve.
Managed Deep Agents allows developers to represent agents as files while offering deep configurability, including custom middleware and custom tools written as code. The service integrates directly with LangSmith Agent Server for streaming, LangSmith Context Hub for context management, and features opinionated memory built on top of Deep Agents. This setup allows practitioners to focus on business logic, instructions, and tools rather than assembling disparate infrastructure components.
This is our own summary of reporting by LangChain Blog
