I've noticed my agent sending key presses to move the screen around and constantly taking screenshots to see if it has made it to the element it wants to view on the page. This is very token intensive sending lots of images as it scrolls down the page looking for something. I often see it do something like:
- Send home key press
- Take screenshot
- Send down arrow key press
- Take screenshot
- Loop 3-4 until it finds what it wants.