Building AI Agents

Agent Architecture

AI Agent Architecture

An AI agent is a system that can perceive its environment, make decisions, and take actions to achieve goals.

Core Components

LLM Core — The "brain" that processes and generates
Memory — Conversation history and context
Tools — Functions the agent can call
Orchestrator — Logic that ties everything together

Agent Loop

1. Receive input
2. Think (LLM generates plan)
3. Act (call tools if needed)
4. Observe (get tool results)
5. Respond (generate final output)
6. Loop back if needed

Design Principles

Single responsibility — Each tool does one thing
Clear interfaces — Well-defined inputs/outputs
Error handling — Graceful failure recovery
Observability — Log important events

Try It Yourself

Run this code example to practice what you've learned.

example.py

1from typing import Callable, Any
2import json
3
4class SimpleAgent:
5    """A minimal AI agent implementation."""
6    
7    def __init__(self, name: str):
8        self.name = name
9        self.tools: dict[str, Callable] = {}
10        self.memory: list[dict] = []
11    
12    def register_tool(self, name: str, func: Callable, description: str):
13        """Register a tool the agent can use."""
14        self.tools[name] = {
15            "function": func,
16            "description": description
17        }
18        print(f"✓ Registered tool: {name}")
19    
20    def think(self, user_input: str) -> dict:
21        """Decide what action to take (simplified)."""
22        # In real agents, this would call an LLM
23        if "calculate" in user_input.lower():
24            return {"action": "calculator", "input": user_input}
25        elif "search" in user_input.lower():
26            return {"action": "search", "input": user_input}
27        return {"action": "respond", "input": user_input}
28    
29    def act(self, action: dict) -> str:
30        """Execute the decided action."""
31        tool_name = action["action"]
32        
33        if tool_name in self.tools:
34            tool = self.tools[tool_name]
35            result = tool["function"](action["input"])
36            return f"Tool result: {result}"
37        
38        return f"I'll help with: {action['input']}"
39    
40    def run(self, user_input: str) -> str:
41        """Main agent loop."""
42        print(f"\n👤 User: {user_input}")
43        
44        # Think
45        action = self.think(user_input)
46        print(f"🤔 Thinking: {action}")
47        
48        # Act
49        result = self.act(action)
50        print(f"🤖 {self.name}: {result}")
51        
52        # Store in memory
53        self.memory.append({"input": user_input, "output": result})
54        
55        return result
56
57# Create agent and register tools
58agent = SimpleAgent("Helper")
59
60agent.register_tool(
61    "calculator",
62    lambda x: "42 (simulated calculation)",
63    "Performs calculations"
64)
65
66agent.register_tool(
67    "search",
68    lambda x: "Found 3 results (simulated)",
69    "Searches the web"
70)
71
72# Run the agent
73agent.run("Can you search for Python tutorials?")
74agent.run("Calculate 6 times 7")
75agent.run("Tell me a joke")