{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/bb7bc8200f54451ab04f287c05a54845\" frameborder=\"0\" width=\"1920\" height=\"1440\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1440,"width":1920,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1440,"thumbnail_width":1920,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/bb7bc8200f54451ab04f287c05a54845-4b53b3a95808aac6.gif","duration":98.343,"title":"How Browser Agents Improve Over Time","description":"This Loom demonstrates how the system routes and verifies browser-based tasks using screenshots and knowledge carried over from previous work. The speaker describes capturing and comparing screenshots, checking whether a company or deficient part is needed, and handling failures based on prior tasks. It also covers an agent harness that self-reports, improves sub-agent performance over time, and includes version history showing which skills were updated and diffs of screen improvements. Finally, the Loom includes a live demo where a browser is spun up and annotated as the page changes."}