Your First AI Agent

Learning Objectives Build a working AI agent from scratch in under 30 minutes Understand the core pattern: model + tools + loop Have a foundation you can extend with more complex capabilities The Simplest Possible Agent Every AI agent, from a simple script to a complex multi-agent system, follows the same core pattern: Receive a task (from a user or another system) Think about what to do (the AI model reasons about the task) Use a tool (execute an action — search the web, run code, read a file) Observe the result (process what the tool returned) Decide: done or continue? (if the task is not complete, go back to step 2) This is the agent loop. Everything else — memory, multi-agent orchestration, complex planning — is built on top of this foundation. Building Your Agent: Three Approaches Approach 1: Claude Code (Fastest Start) If you have Claude Code installed, you already have an agent. Claude Code is an AI agent that can read files, write code, run commands, search the web, and interact with external services through MCP. Try this now: Claude Code will plan the approach, write the code, run it, verify the output, and fix any errors — all autonomously. That is agent behavior: perceive (understand the task), reason (plan the approach), act (write and execute code), observe (check results), and iterate. Approach 2: Anthropic SDK (Python) For building your own agent programmatically, the Anthropic SDK provides tool use out of the box: This is the complete…