LangChain launches Managed Deep Agents service

LangChain has launched Managed Deep Agents, a service that combines the company's Deep Agents harness with managed infrastructure for running agents in production. The release comes roughly a year after Deep Agents first appeared as one of the earlier general-purpose agent harnesses.

The company frames the launch as the latest step in a sequence it traces from early frameworks — LangChain itself, ChatGPT, and AutoGPT in late 2022 — through more mature frameworks like LangGraph and Vercel's AI SDK in 2024, to the point in mid-2025 when models became capable enough to support the now-standard agent pattern: an LLM running in a loop and calling tools. Harness projects like Claude Code and Deep Agents then formalized the tools and environments around that loop.

LangChain says two sets of learnings shaped the new product. First, infrastructure primitives: durable execution to back agent runs, sandboxes for untrusted code, and the common design choice of "separating the brain and hands." Second, emerging standards for controlling harnesses: AGENTS.md for base instructions, MCP for connecting to external systems, and skills for progressive disclosure of context.

Managed Deep Agents bundles the harness with infrastructure rather than leaving builders to assemble the pieces themselves. LangChain lists the bundled components: LangSmith Deployments for runtime, streaming through LangSmith Agent Server and Channels, LangSmith Sandboxes, LangSmith Context Hub for managing instructions and context, Harbor for evaluation, and built-in authentication. Agent definitions are represented as files in a filesystem, and the service permits custom middleware and custom tools written as code.

The product follows two earlier entries in the managed-agent space. Fleet, LangChain's no-code platform, pushed the managed approach furthest, targeting non-developers with a purely UI-based experience — it deliberately represented agents as files, with a toggle to view them in a file explorer. The company says a stream of technical feature requests from developers pushed it toward a more configurable offering. Claude Managed Agents, which launched later, was API-first and targeted developers more directly. Vercel's Eve, released about a month ago, also leans on representing agents as files.

Since this is a company announcement, claims about ease of use are LangChain's own. What is concrete: the product exists, it packages specific components, and it inherits Fleet's file-based agent representation while adding developer-facing configurability.

LangChain also acknowledges the category is unsettled. The company expects more infrastructure requirements and standards to emerge and invites feedback.