Control Flow vs Data Flow

In spaghetti code, data logic and control logic are mixed.

Code Comparison

Bad (Mixed Logic):

# Bad: Node decides where to go next inside its own logic
def my_node(input):
    result = llm(input)
    if "error" in result:
        return run_retry_node(result) # Tightly coupled
    return run_success_node(result)

Good (LangGraph):

# Good: Node only cares about data
def my_node(state):
    result = llm(state['input'])
    return {"output": result}

# Edge handles control flow
def route(state):
    if "error" in state['output']:
        return "retry"
    return "success"

Why Separating Them Matters

1. Reusability: You can reuse the "SearchNode" in 5 different graphs.
2. Visual Clarity: The graph structure defines the logic, not hidden if-statements.