Alibaba’s MAI-UI team published Qwen-UI-Agent, a GUI agent designed to operate mobile apps, desktop software and websites while carrying stateful workflows across devices.
The system combines on-screen actions with CLI and API tools, including batched actions. For builders, that could extend automation into services with limited or missing APIs while keeping longer tasks connected across phones and computers.
On MobileWorld-Real, the team reports a 92.2% success rate across 409 human-written tasks spanning 104 live Android apps. The project says those tasks and trajectories were held out from training, but the result is vendor-reported and has not been broadly validated.
Payments, bookings and other consequential actions still require user confirmation. Production deployments will need similar approval gates, plus their own real-device testing before trusting the agent with sensitive workflows.