Home / Articles / Practical notes: LangChain Deep Agents AI Systems

This article is published in English.

langchain agents llm architecture typescript python

Practical notes: LangChain Deep Agents AI Systems

Operable walkthrough of Practical notes: LangChain Deep Agents AI Systems: contracts, checks, and drop-in code slots for teams shipping this pattern.

1956 words

Use this as an operator-facing rebuild of the ideas in “LangChain Deep Agents AI Systems”: clear stages, ordered code slots, and recovery notes that survive a handoff. The Overview stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

What Deep Agents Actually Is

For the What Deep Agents Actually stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

from deepagents import create_deep_agent

def get_weather(city: str) -> str:
    """Get weather for a given city."""
    return f"It's always sunny in {city}!"
agent = create_deep_agent(
    tools=[get_weather],
    system_prompt="You are a helpful assistant",
)
agent.invoke(
    {"messages": [{"role": "user", "content": "What is the weather in Mumbai?"}]}
)

The Five Capabilities That Make This Different

For the The Five Capabilities That stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

1. Built-in Planning with `write_todos`

For the 1 Built-in Planning with stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness. For the 1 Built-in Planning with stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

2. A Virtual Filesystem

When working through the 2 A Virtual Filesystem stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

3. Subagent Spawning

When working through the 3 Subagent Spawning stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

from deepagents import create_deep_agent, Subagent

code_reviewer = Subagent(
    name="code-reviewer",
    system_prompt="You are an expert code reviewer. Analyze code for bugs, style, and performance.",
    tools=[read_file_tool],
)
agent = create_deep_agent(
    tools=[internet_search],
    subagents=[code_reviewer],
    system_prompt="You are a research and engineering assistant.",
)

4. Automatic Context Compression and Summarization

When working through the 4 Automatic Context Compression stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node. When working through the 4 Automatic Context Compression stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

5. Long-term Memory Across Conversations

The 5 Long-term Memory Across stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

from deepagents import create_deep_agent
from deepagents.backends import CompositeBackend, StateBackend, StoreBackend
from langgraph.store.memory import InMemoryStore

store = InMemoryStore()
backend = CompositeBackend(
    routes={"/memories/": StoreBackend(store=store)},
    default=StateBackend(),
)
agent = create_deep_agent(
    tools=[...],
    backend=backend,
    memory=["path/to/AGENTS.md"],  # persistent context file
)

How It Fits in the LangChain Ecosystem

The How It Fits in stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

Building a Research Agent: The Real Quickstart

The Building a Research Agent stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts. The Building a Research Agent stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

import os
from typing import Literal
from tavily import TavilyClient
from deepagents import create_deep_agent

tavily_client = TavilyClient(api_key=os.environ["TAVILY_API_KEY"])
def internet_search(
    query: str,
    max_results: int = 5,
    topic: Literal["general", "news", "finance"] = "general",
    include_raw_content: bool = False,
):
    """Run a web search and return results."""
    return tavily_client.search(
        query,
        max_results=max_results,
        include_raw_content=include_raw_content,
        topic=topic,
    )
research_instructions = """You are an expert researcher.
Your job is to conduct thorough research and then write a polished report.
Use internet_search to gather information.
Write your findings to files as you go to avoid losing context.
Use write_todos to plan your research steps before starting.
"""
agent = create_deep_agent(
    model="anthropic:claude-sonnet-4-6",  # default model
    tools=[internet_search],
    system_prompt=research_instructions,
)
result = agent.invoke({
    "messages": [{
        "role": "user",
        "content": "Research the current state of agentic AI frameworks in 2025 and write a structured report."
    }]
})
print(result["messages"][-1].content)

The Deep Agents CLI

For the The Deep Agents CLI stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

pip install deepagents
deepagents  # launches the interactive CLI agent

When You Should (and Shouldn’t) Use It

For the When You Should and stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

Why This Matters More Than Another Framework Release

For the Why This Matters More stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness. For the Why This Matters More stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

Getting Started

When working through the Getting Started stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

pip install deepagents tavily-python

export ANTHROPIC_API_KEY="your-key"
export TAVILY_API_KEY="your-key"
export LANGSMITH_TRACING=true  # optional, for debugging
export LANGSMITH_API_KEY="your-key"

Operational checklist

When working through the Operational checklist stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest.

Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

Pin dependency versions and record the image digest that ran the demo. Reproducibility beats tribal knowledge.

Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.