Agentic AI & Automation

Everything you’ve used so far is a chat AI — you say something, it says something back. An agent is different: you give it a goal, and it does things to get there. It runs code, reads files, clicks buttons, looks things up, checks its own work, and keeps going.

The four things that make an agent

• A goal. “Find me flights to Cleveland under $200.”
• Tools. A web browser, a terminal, a calendar, a file system.
• A loop. Try something → check if it worked → try again.
• A way to stop. Done, or out of budget, or stuck.

Why agents are a big deal

A chat AI is a very smart friend you can ask for advice. An agent is a very smart assistant who actually does the thing. That’s a bigger change than it sounds.

Cloud agents vs. local agents

Most famous agents — Claude Code, OpenAI Codex, browser agents — run on the AI company’s servers. Fast, powerful, but your data leaves your computer.

Local agents run the whole model on your own laptop, using software like Ollama or LM Studio. Smaller, sometimes slower, but nothing you say leaves your computer. That’s a big deal for privacy.

Case study — OpenClaw

OpenClaw orchestrates several small local models — like Qwen and MiniMax — to handle different parts of a task. One model plans, another writes code, another checks the output. All of them run on your own computer with Ollama.

A simple OpenClaw session — plain English

1. You: “Organize the files in my Downloads folder by type.”
2. Planner model writes a 5-step plan.
3. Coder model writes Python to move files into subfolders.
4. Runner executes the script in a sandbox.
5. Verifier model reads the result and confirms.

The ethics of agents

Agents can do real damage if they’re wrong. They can delete files, send messages, spend money. Three rules that professional agent-builders follow:

• Sandbox first. Never let an agent touch production without a human in the loop.
• Scope the tools. Only give it the minimum tools it needs.
• Log everything. If something goes wrong, you need to know why.

Capstone prep

In your capstone, you’ll design (not necessarily build) an agent for a real problem in your life. Write the four things: goal, tools, loop, stop condition. Then predict three ways it could go wrong.