A 6-part series

Browser Automation, start to finish

Getting data off the web, or getting an agent to act on it, comes down to picking the lightest tool that does the job and moving up only when you hit a wall. This is the map I use, one surface at a time.

The 90-second version

You need to get data off a website, or have an agent do something on one for you. There are many tools for that, from a plain fetch to a full stealth browser, and it's rarely obvious which one your situation calls for.

The answer is almost always the lightest one that works. Most public pages give up their data to a plain fetch. A real browser earns its place when the page hides behind JavaScript or a login, and stealth tooling only when a site is actively fighting you off. Start simple, escalate only when the current tool fails.

This series walks that ladder rung by rung: the decision framework, then driving a real browser you're already logged into, letting an agent explore a site and then baking the flow into a script, acting inside your everyday browser while you watch, throwaway browsers for tests and screenshots, and what to do when a site is built to block you.

The series

Read in order, or jump to any part.

Get the skill

A sanitized version of the skill I use to pick and drive each of these surfaces, free to download.

browser-automation

View skill →

One front door for every surface in this series. It holds the decision guide, the attended-vs-unattended safety rule, and the connect-and-drive patterns, with the machine-specific parts stripped out so you can wire in your own.