Skip to main content

LangGraph-Style State Machines

Beginner explanation​

A state machine breaks an agent workflow into explicit steps and transitions. That makes the system understandable instead of magical.

Production explanation​

State machines matter when a workflow spans multiple tool calls, retries, approvals, or recoverable failure paths. They give you resumability, traceability, and cleaner separation between planning and execution.

Real-world enterprise example​

A procurement assistant receives a request, gathers vendor context, checks policy, drafts a recommendation, asks for approval, and only then creates the request record.

Mermaid diagram​

TypeScript example​

export interface AgentState {
  userMessage: string;
  evidence: string[];
  approvalRequired: boolean;
  approved: boolean;
  finalAnswer?: string;
}

Python example​

from dataclasses import dataclass, field

@dataclass
class AgentState:
    user_message: str
    evidence: list[str] = field(default_factory=list)
    approval_required: bool = False
    approved: bool = False
    final_answer: str | None = None

Common mistakes​

  • skipping explicit state and hiding everything in messages
  • not modeling rejection or revision loops
  • treating approval as a side effect instead of a transition
  • making retries implicit and therefore hard to debug

Mini exercise​

Model a five-step workflow with one error path and one approval path. List the transition rules before writing any code.

Project assignment​

Use an explicit state definition and node map in Project: Agent Workflow Orchestrator.

Interview questions​

  • What belongs in workflow state and what belongs in transient prompt context?
  • When does a graph provide more value than a simple chain?
  • How do you resume a partially completed run safely?

Monetization angle​

Explicit orchestration is valuable because enterprises want AI automation that can be inspected, approved, and resumed without guesswork.