AI Navigate

Getting Started with AI Agents

Dev.to / 3/20/2026

💬 OpinionTools & Practical Usage

Key Points

  • An AI agent is defined as a system that uses a language model to decide what actions to take, executes those actions via tools, and observes the results to determine next steps.
  • The agent loop is Perceive → Reason → Act → Observe → Repeat, effectively a while loop with an LLM inside it.
  • The guide emphasizes a framework-free, under-one-hour path to a working agent, with a runnable code example and no installations required.
  • It lists framework options (Claude API, LangChain, AutoGen, CrewAI) and provides a quickstart example using Claude, including a Python code snippet.

This article was originally published on do-nothing.ai. The canonical version is there.

Getting Started with AI Agents

Everyone is talking about AI agents. Most of them have never built one.

This guide gets you from zero to a working agent in under an hour. No frameworks to install. No courses to finish. Just the loop that makes agents work, and a code example you can run right now.

An AI agent is a system that uses a language model to decide what actions to take, executes those actions via tools, and observes the results to determine next steps. That is it.

Core Concepts

The agent loop: Perceive → Reason → Act → Observe → Repeat. It is a while loop with an LLM inside it. The industry spent a year naming this.

Tools: Functions the agent can call. Examples: web search, code execution, database queries, API calls.

Memory: How the agent retains context across steps. Short-term (within a conversation), long-term (vector stores or databases).

Framework Options

Framework Language Best For
Claude API (native) Python / JS Simple tool use, direct control
LangChain Python / JS Complex chains, many integrations
AutoGen Python Multi-agent conversations
CrewAI Python Role-based multi-agent teams

Quickstart with Claude

import anthropic

client = anthropic.Anthropic()

tools = [
    {
        "name": "get_weather",
        "description": "Get the current weather for a location",
        "input_schema": {
            "type": "object",
            "properties": {
                "location": {"type": "string"}
            },
            "required": ["location"]
        }
    }
]

response = client.messages.create(
    model="claude-opus-4-6",
    max_tokens=1024,
    tools=tools,
    messages=[{"role": "user", "content": "What's the weather in Paris?"}]
)

What to Build Next

You now have an agent that does exactly one thing. That is more than most people who talk about agents at conferences can say.

  • Give it a second tool and let it choose which one to use
  • Add a loop that feeds tool results back to the model until it says "done"
  • Hook it up to something real: your email, your CRM, your content queue

When you are ready to hand actual business functions to agents, start with How to Delegate Tasks to AI Agents.