Runs: invented users, in a real browser, in real Claude.
An agent plays one invented person, drives a real browser on the live site, tries every example in real Claude, and sends back its screenshots, its report and a record of every step. Each run is replayed here, step by step, with what we found and what we changed.
Run one yourself
- The brief for browser agents: what to give an agent that can drive a browser, such as Claude in Chrome: who to be, the rules, what to capture at every step, and the record to send back.
- The personas: the invented people it may play, and which have been played.
- Study S1: the first study, in a headless browser on a copy of the site, before any run in a real one.
