What do you want to automate today?

This website teaches you what AI agents do, shows how they work via some sample workflows that do sophisticated things, and then gives you a step-by-step method to install an agent (Google's Antigravity) and start automating.

What is an AI Agent?

An AI agent is a thing that controls your computer exactly as a human working on a computer would. It can use applications, browse the web, write files, run code and break massive problems out into smaller steps. It's basically your robot employee.

An ideal agent should

  • Full file access: Can read and write files on your computer.
  • Works on its own: Whether you give it a detailed step-by-step instruction guide (workflow) or you just one-shot a request, it can work for hours without human intervention.
  • Recursive analysis: Generates an output, looks at it, improves it, and repeats until it's exactly what you wanted.
  • Can browse the Internet: Has its own real browser that you see browsing websites. This browser has persistent cookies so once you log in to a site with it you don't have to log in again. This browser normally sends an "I'm a bot" signal to websites, but in lieu of using the agent's built-in browser, you can use the "Control every app in Windows" method below to use your personal browser, in which case that "I'm a bot" signal will not be sent.
  • Control every app in Windows: It can control any app by taking a picture of your screen and finding out where to send the mouse to click on things or to type things into a box. While this method is slow, for some apps (like your Edge or Chrome browser) it can do this much faster by installing a DOM extension. For native Windows apps, it can also use tools like WinAppDriver to interface with the app directly. As agents become more popular, operating systems will let you control apps this fast way rather than having the AI take a picture of the screen, click somewhere, take another picture of the screen, click somewhere, and so on.
  • Terminal execution: Can use the command prompt or other tools to install software, run programs, and fix its own technical errors, just like an IT professional would.
  • Has a memory: To the extent that you want it to remember things, it can remember any aspect of your work or the projects you're working on for as long as you want by saving that information to files.
  • Multi-engine support: The brain of the agent is something like Claude, Gemini, or ChatGPT. An ideal agent lets you pick the one you want for a particular task and switch between them automatically by embedding a command within your workflows.
  • Chunking Rule: If you submit a big prompt to an AI, it gives you an answer with diluted reasoning. If you break it down into 20 smaller subprompts and submit those one at a time, the reasoning quality skyrockets. An ideal agent should do this automatically.

What are workflows?

A workflow is a detailed step-by-step instruction guide for the agent. It goes through it each step at a time and does them. Think of each step as like one prompt to a traditional chatbot AI, except that instead of you sitting there and feeding it those prompts, the agent does it automatically.

The result: outputs that are dramatically more thorough, accurate, and useful than what you'd get from a chatbot AI response.

Would you like to see some sample workflows?