LangGraph Deep Agents — 2026 Production Infrastructure

The "Deep Agent" Abstraction

In early 2026, LangGraph introduced Deep Agents, a high-level abstraction that dramatically simplifies building long-running, stateful systems.

Why Deep Agents?

Previously, developers had to manually write the graph logic for context compression, subagent spawning, and planning loops. Deep Agents encapsulate these patterns natively via deepagents on top of LangGraph v1.x and LangSmith Deployment (formerly LangGraph Platform):

from deepagents import create_deep_agent
from langgraph.checkpoint.postgres import PostgresSaver
from langgraph.types import interrupt, Command

def execute_wire_transfer(amount: float, recipient: str) -> str:
    """Pause mid-node via dynamic interrupt() until a human approves."""
    decision = interrupt({"action": "wire_transfer", "amount": amount, "to": recipient})
    if decision != "approve":
        return "Transfer aborted by reviewer."
    return f"Transferred USD {amount:,.2f} to {recipient}"

with PostgresSaver.from_conn_string(DB_URI) as checkpointer:
    checkpointer.setup()
    agent = create_deep_agent(
        model="anthropic:claude-opus-5-5",
        tools=[execute_wire_transfer],
        checkpointer=checkpointer,
    )
    cfg = {"configurable": {"thread_id": "wire-42"}}
    agent.invoke({"messages": [("user", "Wire 250000 USD to vendor_9")]}, cfg)
    # Resume after human review on LangSmith Deployment:
    agent.invoke(Command(resume="approve"), cfg)

• Native Planning: The agent automatically uses the write_todos pattern to maintain a persistent plan before executing tools.
• Auto-Compression: When the context window fills up, Deep Agents automatically pause, summarize the history, and inject the summary back into state, preventing "Lost in the Middle" failures.
• Dynamic Spawning: Deep Agents can autonomously spawn sub-agents (e.g., spinning up 5 parallel research agents) and aggregate their results without you having to define a static Fan-Out graph.