The Humane AI Pin is now dead after HP acquired its assets, but Rabbit is still pushing forward. In a recent blog post and video, the company debuted its "generalist android agent," demonstrating an AI system capable of operating apps on a tablet—something Rabbit initially promised its R1 device could do over a year ago. However, the R1 failed to deliver on that promise, and it is still incapable of such capacity.
The new demonstration, built on Rabbit's LAM Playground technology, doesn't use the R1 at all. Instead, engineers type requests into a prompt box on a laptop, which further translates them into actions executed on an Android tablet. The AI agent is shown performing tasks such as looking for a YouTube video, searching for a whiskey cocktail recipe, gathering ingredients, and adding them to a grocery list on Google Keep. It can also download the puzzle game 2048 and learn how to play it, albeit slowly.
While the AI agent typically completes these duties, the execution isn't always flawless. For example, when asked to send a poem via WhatsApp, the text was given one message at a time instead of in a single block. An engineer speculated that a better prompt might have prevented this, but they didn't try to improve the request.
The new AI system emphasizes Rabbit's ongoing struggle to fulfill its initial aim. Since debuting the R1 in early 2024 with limited capability, the business has been regularly releasing improvements, such as training AI agents for specialized jobs and customizing the user interface. According to Rabbit's blog, the most recent presentation only shows the Android agent's "core action loop." The company promises to reveal more details about its "upcoming cross-platform multi-agent system" in the coming weeks.
The Rabbit's latest showcase is a step forward; it raises concerns about the company's ability to realize its ambitions completely or if it's still trying to catch up to what it initially promised.