i had this idea early this year to build github copilot for the browser not an "agent" but just a next-click predictor/text autocomplete based on accumulated context↗