Rabbit reveals Android agent that can control your phone

Rabbit just offered a research preview of an autonomous Android agent capable of controlling mobile apps and navigating system-level tasks on a user's behalf.

Following its web-focused LAM playground launched last October, the company is shifting gears directly into mobile operating systems as part of its goal to build a cross-platform generalist agent.

Driving apps across the OS

In video demonstrations released alongside the announcement, the Android agent handled both straightforward system tweaks and multi-step workflows across separate apps.

On the simple end, the agent navigated through Android system settings to toggle notification preferences for a specific app.

It also tackled routine social and entertainment requests, such as generating a poem with AI and messaging it via WhatsApp, or searching for and playing a specific video inside the YouTube app.

Where it got more ambitious was chaining tasks across different services without human intervention.

The agent pulled a cocktail recipe from a dedicated app and organized the ingredients into a grocery list inside Google Keep. In another workflow, it retrieved a business idea from Keep, drafted a revenue projection within Google Docs, and shared the finished file with contacts.

In a gaming demo, it even navigated Google Play, installed a game, and launched directly into playing it.

Rabbit noted that the preview primarily demonstrates the core action loop of how the agent reads the screen and acts. The team acknowledged that significant work remains around execution speed, general intelligence, and how fast the agent recognizes UI states, promising performance updates in the coming weeks alongside details on a broader multi-agent framework.

Why it matters

For smartphone users, an agent that actually handles mundane mobile chores has long been the holy grail, but we think it is far too early to rely on this for daily use. While chaining Keep to Docs looks neat in controlled clips, real-world Android apps break constantly with dynamic UI updates, pop-ups, and authentication walls. Until safety and accuracy benchmarks are proven, having an experimental agent touch work documents or payment flows remains high risk.

For the broader industry, Rabbit is pivoting directly into the territory where Google itself is racing with native Gemini integration. Building an overlay agent on Android without native OS control is technically fragile and expensive to run at scale.

Rabbit is making a smart strategic pivot away from bespoke hardware toward pure software agency. However, the catch is glaring: third-party screen-scraping agents will always struggle with latency and fragile permissions compared to first-party mobile operating systems. Unless Rabbit can deliver flawless reliability, competing against native assistants on their home turf looks like an uphill battle.