AI Agents: From Architecture to Production · Lesson 1

The Agent Loop: Perception → Reasoning → Action

Breaking down the perception → reasoning → action → observation cycle, how LLMs make decisions inside an agent, and why agents are fundamentally different from chatbots.

30 min read2 questions in quizReady prompt includedIn progress

Practical exercise

What to do after this lesson

Implement a minimal agent loop: the agent receives a user question, decides whether search is needed, runs it (mock function), and returns an answer. Add an iteration counter with a limit of 5.

Ready-to-use prompt

Template for this lesson

Copy and adapt to your context. Text in angle brackets should be replaced.

You are an AI agent that answers questions using web search.
Task: <user question>

Think step by step:
1. Is a search needed to answer this?
2. If yes — formulate a search query.
3. After getting results — give the final answer.

Common mistakes

What people get wrong

No iteration limit — agent loops forever on tool errors.
Not appending tool_use results to messages — the model doesn't see what happened.
Confusing stop_reason — continuing after end_turn.

Pro tips

What works but no one documents

Log every loop iteration — it makes agent behavior much easier to debug.

The Agent Loop

Each iteration has four steps:

Perception → Reasoning → Action → Observation ↑ │ └─────────────────────────────────────────┘

| Step | What happens | |------|-------------| | Perception | The agent receives inputs: user request, previous tool results, memory contents | | Reasoning | The LLM decides: answer in text, call a tool, or finish | | Action | A tool runs (your code), an API is called, a file is written | | Observation | The action result is added to context — the loop starts again |

How the LLM Makes Decisions

The model doesn't "think" in the intuitive sense. It generates the next token based on the entire context — system prompt, conversation history, tool descriptions, and previous results.

Three possible outcomes each iteration:

Final answer — the model considers the task done.

Tool call — the model generates a JSON request instead of text.

Chain-of-thought — the model reasons aloud before deciding.

import anthropic client = anthropic.Anthropic() tools = [ { "name": "search_web", "description": "Search the web for current information. Use when the user asks about recent events or facts you may not know.", "input_schema": { "type": "object", "properties": { "query": {"type": "string", "description": "Search query"} }, "required": ["query"] } } ] def run_agent(user_message: str) -> str: messages = [{"role": "user", "content": user_message}] for _ in range(10): # iteration limit — mandatory safety guard response = client.messages.create( model="claude-opus-4-5", max_tokens=1024, tools=tools, messages=messages, ) if response.stop_reason == "end_turn": return response.content[0].text if response.stop_reason == "tool_use": messages.append({"role": "assistant", "content": response.content}) tool_results = [] for block in response.content: if block.type == "tool_use": result = execute_tool(block.name, block.input) tool_results.append({ "type": "tool_result", "tool_use_id": block.id, "content": result, }) messages.append({"role": "user", "content": tool_results}) return "Max iterations reached" def execute_tool(name: str, input: dict) -> str: if name == "search_web": return f"[Search results for '{input['query']}': ...]" return "Tool not found"

Report a bug

The Agent Loop: Perception → Reasoning → Action

Quiz — 2 questions

What Is an AI Agent

The Agent Loop

How the LLM Makes Decisions

Agents vs Chatbots

When the Loop Stops