We're sorry but this app doesn't work properly without JavaScript enabled. Please enable it to continue.

This lesson's interactive features are locked, please to keep using them

Browser Automation

Now, a browser is still a graphical application, but it does expose way more structure than your typical native desktop app. It makes sense: web apps are built on structured text! So text-based browser automation tools can easily inspect:

  • The document structure
  • The accessibility tree
  • Console messages
  • Network requests
  • Page layout measurements

That means a text-first agent like OpenCode can still fill the Company field by name, click the Add application button, and then check that the new row appeared. It doesn't even need an image model to guess where the button's pixels happen to be.

Browser Feedback Loops

OpenCode can drive a browser automatically to run a full validation loop, like this:

  1. Start the application.
  2. Open the page in a (usually headless) browser.
  3. Reproduce the bug with the same clicks a user would make.
  4. Inspect the DOM, developer console, and network activity.
  5. Fix the code and run the flow again.

This works incredibly well for exploration and debugging – and you'll get much better results than if you just had the model one-shot a solution before you checked it manually. That said, you still do need to check it manually.

Vision-enabled models that can actually use the computer will catch more visual bugs than a text-only model, but they won't catch everything, and they still lack taste.