AI Can Now Use Any Software — No API Required
Alibaba's new Qwen-UI-Agent can operate phones, computers, and websites just like a person does, and it already outperforms rival AI from OpenAI and Anthropic.
- 82.1%
- MobileWorld benchmark score
- 12 points
- ahead of OpenAI GPT-5.6
- 100+
- phones in training data
- Aug 20
- release date
Most business automation runs through APIs — the hidden links that let apps talk to each other. If a tool has no API, there is no easy way in. That barrier is starting to fall. A new kind of AI — called a GUI agent — can look at a screen and operate software exactly as a person would: clicking, typing, and navigating step by step.
On August 20, Alibaba released Qwen-UI-Agent. It can control phones, computers, websites, and command-line tools — all in the same session. On the MobileWorld test, which measures real-device navigation, it outscored every rival model tested, including GPT-5.6 (by 12 points) and Claude Opus 4.8 (by nearly 15). The model was trained on data from more than 100 phones and 150 apps.
The practical change this unlocks is significant. Any software a person can use, an AI agent can now potentially run. Legacy tools with no API, niche industry software, even ordinary websites all become candidates for automation. A finance team could automate work in old spreadsheet software. A law firm could pull data from a case management tool. An e-commerce team could run competitor price checks across any site without a special integration.
If you are building or running something that depends on software connections, this is worth a test right now. Pick one workflow that still needs a human to click through it. Try a screen-aware agent on it. The list of automatable tasks just got much longer — and starting is simpler than building a custom integration from scratch.