<< back to Guides

AI Core Concepts (Part 14): AI Agents

AI Agents are autonomous systems that can perceive their environment, make decisions, and take actions toward achieving goals. Unlike simple models, agents often act in a loop, learn from outcomes, and adjust behavior over time.

1. What Is an AI Agent?

An AI Agent typically follows the perception → decision → action cycle.

Perception: Observes data from the environment.
Decision-making: Plans what to do using a model or policy.
Action: Executes steps in the environment.
Feedback Loop: Observes effects and learns from them.

2. Types of AI Agents

Agent Type	Description
Reactive Agent	Responds to current input (no memory/learning)
Reflex Agent	Uses rules to map inputs to actions
Goal-Based Agent	Chooses actions to achieve specific goals
Learning Agent	Improves over time using data/experience
Planning Agent	Builds plans by simulating outcomes

3. Basic AI Agent Structure (Pseudocode)

while True:
    observation = agent.observe(env)
    action = agent.decide(observation)
    env.update(action)
    agent.learn(observation, action, env.feedback())

4. AI Agents vs Standard Models

Feature	Standard ML Model	AI Agent
Passive or Active	Passive (predict-only)	Active (acts in environment)
One-shot inference	Yes	No, typically multi-step
Learning approach	Supervised, Unsupervised	Often Reinforcement Learning
Autonomy	Low	High

5. Example Use Cases

Robotics – Robots navigating a room, picking objects
Game Agents – AI that learns to play video games (e.g., AlphaStar, MuZero)
Conversational Agents – Multi-turn chatbots with memory and reasoning
Automation Bots – Web scraping agents, workflow automation (RPA)
Trading Bots – Agents making buy/sell decisions in markets

6. Example: Agent with LangChain + OpenAI

from langchain.agents import initialize_agent, Tool
from langchain.llms import OpenAI
from langchain.tools import DuckDuckGoSearchRun

search = DuckDuckGoSearchRun()
tools = [Tool(name="Search", func=search.run, description="Useful for answering factual questions")]

agent = initialize_agent(
    tools,
    llm=OpenAI(temperature=0),
    agent="zero-shot-react-description",
    verbose=True
)

agent.run("What's the weather in Berlin today?")

7. Memory & Planning (Advanced Agents)

Agents can use memory to store conversation history or past observations.

Short-term memory: Limited context window (chat context)
Long-term memory: Vector store or database
Planning: Create multi-step strategies (e.g., AutoGPT)

# Pseudocode for a planning agent
goal = "Book a flight and hotel to Paris"
plan = planner.decompose(goal)
for step in plan:
    agent.execute(step)

8. Frameworks for Building AI Agents

LangChain – Chains of LLMs, tools, memory, and agents
AutoGen by Microsoft – Agents collaborating via LLMs
Haystack – Conversational pipelines with retrievers
CrewAI – Define multi-agent collaborations
AgentLite / Semantic Kernel – Lightweight frameworks for agent workflows

📚 Further Resources

<< back to Guides