Honor is pushing deeper into agentic AI with Honor UI Agent, a phone assistant built to read what is on screen and act through app interfaces. The company presented the system around a familiar task: booking a restaurant table through OpenTable.
What Honor UI Agent Is Trying To Do
Honor describes Honor UI Agent as a "GUI-based mobile AI agent." In plain terms, the assistant is meant to understand the graphical user interface on a phone screen, then use that understanding to complete tasks for the user.
WIRED saw an early demo before Honor's keynote at Mobile World Congress 2025 in Barcelona. At the event, Honor also announced its $10 billion Honor Alpha Plan, a long-term effort tied to AI development across the company's personal technology devices.
The plan was envisioned by Honor's new CEO Jian Li. Its stated goals include "creating an intelligent phone" and "open human potential boundaries and cocreate a new paradigm for civilization." Behind that corporate language is a simpler signal: Honor is making AI a central part of its device strategy.
The Restaurant Demo Showed Both Progress And Limits
In the demonstration, an Honor spokesperson asked the agent to book a table for four people, gave a time, and requested "local food." Because the demo took place in Barcelona, the agent treated that as Spanish food.
The assistant then moved through the OpenTable app to search for a restaurant and attempt the booking. The process was visible on the phone screen, which made the automation feel less invisible than the idea of an AI assistant might suggest.
Honor told WIRED that, in the future, its UI Agent will not need to show its work in the same way. For now, the demo made the mechanics clear: the agent was not simply answering a question; it was interacting with the app interface step by step.
The agent did not complete the booking. It selected a restaurant, but the reservation required a credit card to confirm, so the user had to take over.
Honor also showed that prompts can be flexible. In another example, asking for a "highly rated" restaurant led the agent to look at reviews with high scores. The source notes that it did not go beyond that by comparing OpenTable reviews with information from other parts of the web.
Why The Screen-Based Approach Matters
Honor's idea sits inside the wider push toward agentic artificial intelligence: systems that do not just respond, but take actions on a user's behalf. The source article places Honor UI Agent alongside other efforts, including an AI assistant tested by Will Knight that could browse the web and perform online tasks, and Google's Gemini 2 AI model, which was unveiled late last year and trained to take actions for users.
The approach also connects to the idea of a generative user interface for smartphones. At MWC 2024, some companies were exploring ways to interact with apps without using apps directly, leaning instead on AI assistants that generate a user interface after a command.
Honor's method is different from relying only on app APIs. The source compares it with Rabbit's Teach Mode for the Rabbit R1, where a user manually trains the assistant to complete a task. Honor says its system is not trained to follow strict steps in that way.
Instead, Honor says the agent uses multimodal screen context recognition. The claim is that it can understand semantic elements in the user interface and carry out a multi-step request without the user teaching it every part of the app.
Honor also argues this is more cost effective. The company said: "Unlike competitors such as Apple, Samsung, and Google, which rely on external APIs—resulting in higher operational costs—Honor's AI Agent independently manages a wide range of tasks."
Gemini 2, On-Device Data, And Personal Context
Honor says the UI agent uses in-house execution models. It also uses Google's Gemini 2 large language model for intent recognition and for "enhanced semantic understanding" of what appears on the screen.
Google did not share details about the nature of the collaboration. That leaves an important part of the system unexplained from the outside: how Honor's execution models and Gemini 2 divide the work in practice.
Honor also says it has partnered with Qualcomm to keep data on the device and to develop a personal knowledge base that learns preferences over time. The example given is food delivery: if a user tends to order certain kinds of food, the agent could use that context when asked to order on the user's behalf.
The company says it is already employing some of these AI agents in China. Based on the demo, the main promise is convenience inside familiar apps. The main caution is equally clear: when a task reaches sensitive steps, such as confirming a reservation with a credit card, the human still matters.
Honor's Broader Device Push
At the same keynote, Honor announced seven years of software updates for its flagship Magic 7 Pro and upcoming devices. The source notes that this matches the software update policies from Google and Samsung for Pixel and Galaxy phones.
Honor also unveiled several products at the show: Honor Earbuds Open, Honor Watch 5 Ultra smartwatch, Honor Pad V9 tablet, and Honor MagicBook Pro 14 laptop.
Those devices will not be sold in the US, like most Honor products, but they will be available in other markets. For Honor, the broader message at Mobile World Congress 2025 was that AI is not being treated as a single feature. It is becoming a layer across phones, apps, and personal devices.