Rabbit has recently showcased its new AI agent despite the collapse of its Humane AI Pin. In a blog post and video, the company revealed a “generalist Android agent” that demonstrates how it can control apps on a tablet, similar to what Rabbit had promised with its R1 device over a year ago. However, the R1 device itself is not used in this demonstration.
Instead, the engineers input requests into a prompt box on a laptop, which translates these commands into actions on an Android tablet. For example, they ask the agent to find a YouTube video, locate a whiskey cocktail recipe in an app, gather the necessary ingredients, and add them to a Google Keep grocery list. At one point, they also request the agent to download the puzzle game 2048 and learn how to play it, which it manages to do, albeit at a slow pace.
The AI agent performs the tasks with varying degrees of success, sometimes executing requests well but also exhibiting quirks, such as sending a poem over WhatsApp one line at a time instead of in a single block. One engineer even speculates whether they should have included line breaks in their prompt but decides against trying again.
This showcases that Rabbit’s AI agent is still a work in progress, reflecting the challenges it has faced since the R1’s launch, which fell short of the capabilities promised by CEO Jesse Lyu in January 2024.
Despite these challenges, Rabbit has consistently rolled out updates, including features that allow users to train their AI agents to perform specific tasks or modify their interface. The company indicates that the examples shared in the recent demonstration represent just the basic functions of an Android agent. Rabbit plans to provide more information about its upcoming cross-platform multi-agent system in the near future.
