9.1.1 Agent Basics Roadmap: Goal, State, Action

An Agent is not a model name. It is a system pattern that uses a model, tools, state, memory, and feedback to keep working toward a goal.

See the Single-Agent Loop First

A normal chatbot answers once. A workflow follows fixed steps. An Agent can plan, act, observe, update state, and continue when the goal is not done.

Run a Tiny Agent State Loop

This script does not call a model yet. It shows the minimum state you need before an Agent can be debugged.

goal = "summarize RAG citation rules"
state = {"steps": [], "done": False}

for action in ["plan", "search_docs", "summarize"]:
    state["steps"].append(action)

state["done"] = True

print("goal:", goal)
print("steps:", " -> ".join(state["steps"]))
print("done:", state["done"])

Expected output:

goal: summarize RAG citation rules
steps: plan -> search_docs -> summarize
done: True

If a demo cannot show goal, state, action, observation, and stop condition, call it an LLM app first, not an Agent.

Learn in This Order

Step	Read	Practice Output
1	What is an Agent	Compare chatbot, workflow, RAG app, and Agent
2	Development history	Understand why LLMs revived Agent systems
3	Capability levels	Place answer, retrieve, tool use, plan, memory, collaboration on one ladder
4	System architecture	Draw goal, state, planner, tools, memory, observation, executor
5	RL to Agent breakthroughs	Connect action, reward, feedback, and planning

Evidence to Keep

Keep this page’s proof of learning as a small evidence card:

Agent Boundary

how this differs from chatbot or fixed workflow

Goal State Action

goal, current state, next action, observation

Architecture Parts

planner, tools, memory, guardrails, evaluator

Failure Check

over-autonomy, vague goal, missing state, or no trace

Next Action

build the smallest traceable single-agent loop

Pass Check

You pass this chapter when you can draw one single-Agent loop and explain why single-Agent stability comes before multi-Agent collaboration.

The exit mini project is a research assistant Agent trace: one goal, one plan, at least one tool decision, one observation, one stop condition, and one final answer.

Check reasoning and explanation

1. A passing answer describes the agent loop: goal, plan, tool call, observation, memory or state update, and stop condition.
2. The evidence should include a trace that another developer can inspect, not only the final answer.
3. A good self-check names one safety or reliability control such as tool schemas, permission boundaries, retries, evaluation cases, or a human-review point.