Your Phone Can Do the Work Now: Zhipu AI’s AutoGLM Opens the Door to Fully Automated Smartphones
Imagine telling your phone, “Book me a flight to Tokyo, get me a latte at 8am, and message my friend I’m on the way,” and the phone does it — taps, swipes, types, browses, and even hits “buy.” This is no sci-fi fantasy. On December 9, 2025, Chinese AI startup Z.ai (formerly Zhipu AI) released into the wild its open-source “phone agent” framework: AutoGLM. (Tech in Asia)
What AutoGLM Does — and Why It’s a Big Deal
- AutoGLM is built to operate a smartphone like a human: it interprets screen content and simulates taps, swipes, typing — essentially controlling a phone UI. (iThome)
- It supports complex, multi-step workflows: food delivery orders, flight bookings, purchases, messaging etc. — anything that might take dozens of steps on your smartphone. (Asia Business Outlook)
- The initial release supports more than 50 popular Chinese apps, including WeChat, Taobao, Douyin, Meituan etc. (TechNode)
- The open-source package includes everything developers need: trained models, adaptation layers for Android, a full “phone-use” framework, toolchain, runnable demos, and documentation. (TechNode)
- Importantly, Z.ai says AutoGLM can be deployed locally (on device) or in the cloud, offering flexibility while preserving user privacy and data control. (Asia Business Outlook)
In short: for the first time, a publicly available AI agent can (at least in principle) automate everyday smartphone tasks — the very essence of turning your phone into a fully autonomous digital assistant.
Why Now? And What It Says About the AI Race
Z.ai’s release of AutoGLM comes at a pivotal moment for China’s AI landscape:
- Backed by previous successes like the open-source foundation model GLM‑4.5 (launched earlier in 2025), Z.ai has shown ambition to push beyond just chatbots. (Computerworld)
- The open-sourcing of AutoGLM — under what the company calls a “public foundation collectively owned and fine-tuned by the whole industry” — signals a shift from walled-garden AI experiences to community-driven ecosystems. (South China Morning Post)
- Z.ai itself argues that “AI phone capability” shouldn’t be monopolized by a few players: broader access enables more innovation. (Sina Finance)
In effect, AutoGLM could serve as a foundation for third-party developers, device makers, and even hobbyists to build their own “smartphone AI assistants.” Over time, this may drive a wave of new AI-powered smartphone experiences — beyond what most manufacturers ship today.
But It’s Not Magic — Yet
AutoGLM’s promise is huge — but the reality comes with caveats. According to one preliminary assessment in a Chinese tech outlet: while AutoGLM impressed in demos (e.g. bulk-sending red packets in a chat group), real-world usage often faltered. Variations in UI layout, different phone types (foldable screens, different resolutions), or unpredictable pop-ups (ads, permission dialogues) all undermined reliability. (Huxiu)
Moreover, while AutoGLM is open source, licensing details remain unclear. Some reporting warns that missing or ambiguous licensing may limit commercial use — a potential barrier for hardware makers or mainstream app integration. (Asia Business Outlook)
Finally: current performance isn’t perfect. Phone-use AI requires extremely robust perception (detecting UI elements) and precise control — a notoriously hard problem in real-world app environments. AutoGLM is a big first step, but likely not ready to run 100% reliably for every user.
What This Means — And What’s Next
- AutoGLM could accelerate AI-powered phone assistants, where your phone isn’t just a tool, but an active agent doing things on your behalf.
- If widely adopted, it may spark new innovation: third-party AI assistants, custom automation workflows, and more personalized “phone agents.”
- It may also raise questions around privacy, security, and trust — especially if AI agents are granted deep control over personal apps, messaging, purchases, and more.
- For AI developers, AutoGLM could become a testing ground: building and refining models to handle the messy, unpredictable world of real smartphone use.
If AutoGLM lives up to its promise — and if its limitations are addressed — the concept of a “do-everything smart assistant” inside your smartphone might finally shift from sci-fi hype to everyday reality.
Glossary
- AI Agent: A software system that can interact with its environment, perceive state, plan and execute actions — more than just responding to user prompts.
- Open source: Code (and sometimes models) that is publicly released, so others can view, use, modify and distribute it.
- LLM (Large Language Model): A deep learning model trained on massive amounts of text data, which can understand and generate human-like language.
- Phone-use framework / GUI agent: A system that enables an AI to control a smartphone’s user interface — simulating taps, swipes, typing, and reading on-screen content.
Source: https://www.techinasia.com/news/chinas-zhipu-ai-launches-open-source-phone-agent
