DeepAgents: How LangChain's New Framework Makes AI Actually Do the Work

Forum topic · 小凯 · 2026-03-23

Summary

Most AI assistants today behave like a knowledgeable advisor: they explain what should be done but never execute the task themselves. DeepAgents, a new open-source framework from the LangChain team, aims to close that gap by turning conversational AI into agents that plan and act independently. The framework is built on four pillars: a planning tool that auto-generates and tracks a to-do list for each task; a built-in filesystem with full file operations so the agent can organize, save, and edit files across a workspace; sub-agents that each carry independent context to handle complex subtasks without interference; and context management that auto-summarizes conversations and offloads large outputs to files to prevent long-session memory loss. DeepAgents is explicitly inspired by Anthropic's Claude Code but differs in three key ways: it supports any LLM rather than only Claude models, it is MIT-licensed open source, and it is LangGraph-native, integrating with LangSmith, persistence, and checkpointing. An agent can be created with a single line of code, with all advanced features enabled out of the box.

You may have wondered about this question:

Today's AI can chat, write articles, and even write code — but can it truly complete tasks independently?

For example: you say, "Help me research the EV market and write a report." The AI replies with a long message telling you "which aspects to research, where to get the data, what structure the report should have."

Sounds professional, right? The problem is — you still have to do all the actual work yourself.

That's no surprise. Most AI assistants are essentially a "fancy search engine + text generator": they give advice but don't execute; they provide information but don't act. Like a learned consultant standing by, pointing out strategy but never rolling up their sleeves.

Today's topic might change that.

It's called DeepAgents, a new framework just released by the LangChain team. Its ambition is simple: turn AI from a "parrot that only talks" into an "assistant that actually works".

🛠️ Four Pillars: DeepAgents' Core Weapons

📝 Pillar 1: Planning — The Art of Fitting an Elephant into a Fridge

DeepAgents ships with a built-in planning tool. Once you assign a task, it automatically generates a to-do list and tracks its own progress.

📁 Pillar 2: Filesystem — Giving the AI a Real "Swiss Army Knife"

Complete built-in file operation tools: the agent can create folders, categorize and save materials, and modify multiple files on its own.

🐣 Pillar 3: Sub-agents — Solving the "Not Enough Brainpower" Problem

When a task gets too complex, the main agent can dispatch sub-agents to handle specific subtasks. Each sub-agent has its own independent context, with no interference.

🧠 Pillar 4: Context Management — Giving the AI a "Notepad"

Automatic summarization plus automatic offloading of large outputs to files solve the "amnesia" problem in long conversations.

🎭 Relationship with Claude Code

DeepAgents explicitly says it is deeply inspired by Claude Code, with key differences:

Claude Code only works with Claude models; DeepAgents supports any LLM
Claude Code is closed source; DeepAgents is MIT open source
DeepAgents is LangGraph-native, integrating with LangSmith, persistence, and checkpointing

🔧 Out of the Box

Create an agent with a single line of code — all advanced features (planning, file operations, sub-agents, context management) work out of the box.

📚 Key References

1. LangChain. (2025). DeepAgents. https://github.com/langchain-ai/deepagents 2. ByteIota. (2026). DeepAgents Tutorial. https://byteiota.com/deepagents-tutorial 3. Anthropic. (2025). Claude Code. https://www.anthropic.com/claude-code