Reach new heights of laziness by burning tokens to let an AI tap your smartphone screen for you

Opposable is a tool that lets AI agents operate smartphone screens by tapping, swiping and typing on a user’s behalf. Its launch highlights the trade-off in automating simple phone tasks: the system can be slow, unreliable and costly, even when those tasks are quick to do manually or with a phone’s built-in assistant.

On Android, Opposable uses accessibility services and screenshots, with a vision-language model to work out where to tap; around 40 common actions instead run through fixed code. Its examples took 50 seconds and 17 tool calls to switch to dark mode and set a timer, nearly two minutes to answer an Apollo 11 query at a cost of $0.44, and 96 seconds to open a stopwatch. In tests on eight unfamiliar tasks, it succeeded three or four times out of eight; checking each subgoal raised success to six out of eight in both rounds. iPhones need a companion computer, while Android can also run an on-device agent.