0:00 / 0:00

Building Deep Agents and Deploying in Production

Name: Building Deep Agents and Deploying in Production
Uploaded: 2026-07-31T19:19:14.000Z
Duration: 15 min 40 s
Channel: LangChain
Description: Robert Xu from LangChain explains what Deep Agents are, how the 'harness' concept works, and how to deploy Deep Agents in production with focus on durable execution, memory management, and human-in-the-loop design.

LangChain31 July 2026Watch on YouTube

Part of series

Ep. 7 · Agents Deep Managed

View the series

What you'll learn

Deep Agents combine prompts, memory, tools, and MCP as core primitives and can be integrated into LangGraph workflows with a single line change.
The harness concept provides additional control and reliability through hooks and middleware, with durable execution and checkpoint recovery for production.
Short and long-term memory management, authentication, and human-in-the-loop design with interrupts and approvals are essential for production deployment.
The file system serves as the agent's scratch pad, supporting flexible state and working memory management.

Frequently asked questions

What is the difference between Deep Agents and standard agents?

Deep Agents use a harness concept that provides additional control and reliability through hooks and middleware, making them better suited for complex tasks and production environments. They combine prompts, memory, tools, and MCP as core primitives.

How do you implement human-in-the-loop design in Deep Agents?

Human-in-the-loop is implemented through interrupts, approvals, and streaming, allowing users to review and approve agent actions before execution.

What are the main challenges when deploying Deep Agents to production?

The main challenges are durable execution with failure recovery, short and long-term memory management, authentication and RBAC, and implementing human-in-the-loop controls.

How do you integrate Deep Agents into LangGraph workflows?

Deep Agents can be integrated into LangGraph workflows with just a single line change, enabling simple composition of agents within larger agentic workflows.

Topics

deep agents langchain langgraph production deployment agentic ai

Description from the channel

Robert Xu from the LangChain team breaks down what Deep Agents actually are, how the "harness" concept fits into the model plus harness equation, and how Deep Agents relate to LangChain and LangGraph. The second half covers what changes when you move a deep agent into production, including durable execution, short and long term memory, authentication, and human in the loop design. Recorded at a LangChain meetup. Chapters: 0:00 Kicking off: deep agents and going to production 0:27 What is a deep agent, and what's a harness 1:07 The Claude Code example: 500,000 lines of harness 1:40 The core primitives: prompts, memory, tools, MCP 2:25 Hooks and middleware, the extra lever 2:48 Why deep agents ships fully open source 3:26 The file system as your agent's scratch pad 4:18 Where deep agents sits in the LangChain stack 5:43 One line change: turning any agent into a deep agent 6:31 Composing deep agents inside LangGraph workflows 7:16 The reliability vs agency tradeoff curve 8:16 Moving deep agents into production 9:48 Durable execution and recovering from failures 11:00 From checkpoints to short and long term memory 12:15 Authentication, RBAC, and the new auth problem 13:48 Human in the loop: interrupts, approvals, streaming 15:06 The LangSmith deployments pitch