← back to topics

Browser Automation

Driving a real web browser programmatically for testing, scraping, and increasingly for AI agents that navigate the web.

exploring systemstestingagents

Browser automation controls a real browser through code: navigating pages, clicking, typing, waiting for content, and reading the DOM (Document Object Model), without a human driving it.

Why I’m digging in

  • It powers end-to-end testing, scraping, and workflow automation, and it is now the substrate for AI agents that use the web the way a person does.
  • It connects directly to the headless idea from the Nous write-up

What I want to understand

  • The main engines (Playwright, Puppeteer, Selenium) and where each one fits.
  • The Chrome DevTools Protocol (CDP) underneath, and how a headless Chromium is actually driven.
  • Robust selectors and waiting strategies, since timing and flakiness are the hard part of reliable automation.
  • The new wave of AI-driven browser agents (computer use, browser-use) and how they turn a goal into concrete clicks and keystrokes.

Starting points

  • Playwright, especially its auto-waiting model and trace viewer.
  • The Chrome DevTools Protocol as the layer most tools build on.
  • Browser-agent projects that let an LLM drive a page toward a goal.

Notes

  • 2026-07-06: I want to understand auto-waiting and selector robustness deeply, then try an LLM-driven agent that completes a small web task start to finish.