The world just got a glimpse of what the future of smartphones looks like and it fits right in your pocket. At China’s annual World Artificial Intelligence Conference in Shanghai, tech companies unveiled a new generation of AI-powered devices that go far beyond voice assistants. These are full AI agent smartphones, and they are here to change everything about how we use our phones.
Background
For years, the idea of a truly intelligent phone one that could think, plan, and act on your behalf remained in the realm of science fiction. That changed fast. China’s AI boom, led by models like DeepSeek and tools like Doubao, gave developers the raw intelligence needed to build phones that don’t just respond to commands but actually carry them out across different apps. The World Artificial Intelligence Conference in Shanghai this weekend became ground zero for that shift.
The Rise of the Pocket Size AI App Phone
Think about the last time you had to switch between five different apps just to book a dinner reservation. You opened a browser, searched for a restaurant, jumped to a maps app, then a booking platform, then fired off a message to a friend. That whole process, according to companies showcasing at this year’s conference, will soon be handled by a single voice command to your pocket size AI app.
This is not a future concept. It is happening right now, and China is leading the charge. At least three major firms were showing off what the industry is calling “agentic phones” smartphones powered by AI agents that can take real actions across apps, platforms, and services all on their own. The energy at the conference felt different this year. It was less about what AI could theoretically do and more about what it is actually doing, right now, in devices small enough to fit in your jeans pocket.
Nubia NaviX Ultra: The Phone That Thinks for You
One of the biggest announcements came from Chinese smartphone maker Nubia, which unveiled its new NaviX Ultra a device powered by Doubao, the wildly popular AI chatbot from ByteDance, the company behind TikTok. Nubia called it simply: “A new era of AI agent smartphones begins.”
That is not just marketing talk. The NaviX Ultra is designed to let users speak a single command say, “order my usual lunch” and have the phone handle everything else. It finds the app, places the order, and confirms the delivery. No tapping required. No switching between screens. Just talk, and the pocket size AI app does the rest.
This concept was tested earlier with a limited-run prototype called the “Doubao Phone,” which sold out almost immediately after its December launch. The prototype could follow voice commands to execute tasks across apps, including ordering food and comparing shopping prices across multiple platforms. The response from early users was overwhelmingly positive, proving real demand for this kind of AI PMX pocket-based machine execution.
When Big Tech Pushed Back
Not everyone was happy about it. Days after the Doubao Phone launched, tech giants including Alibaba, Tencent, and JD.com moved to block the phone’s built-in AI assistant from accessing their platforms. The message was clear: they were not ready to hand over user interactions to a third-party AI agent.
ByteDance responded by disabling the most powerful features of the agent in certain situations, particularly when payments were involved. It was a significant stumbling block but also a sign of how disruptive this technology really is.
Kiranjeet Kaur, associate research director at IDC, explained the tension well. Platforms want to remain the direct point of contact with their users. The moment an AI agent steps in between, she said, “they lose control to another party.” That is a big deal in an ecosystem where user data and transaction control mean billions of dollars.
“Agenting is everyone’s dream, but we haven’t reached there yet,” Kaur noted, adding that the performance of AI agent tools is still uneven.
According to Chinese tech media, Nubia’s new NaviX Ultra addresses this issue differently. Rather than trying to force its way into apps without permission, it works with app platforms through formal agreements a smarter, more sustainable approach to building an Anwar AI ecosystem where phone intelligence and app ecosystems work together rather than against each other.
Honor’s Robot Phone: AI With a Personality
Another company making waves at the conference was Honor, which brought its so-called “Robot Phone” to the show floor. This device features an interactive camera that physically flips up on a small robotic arm giving the phone a surprisingly lifelike quality.
Honor describes it as a “companion-centric” device. The camera can interpret human gestures, respond to music by moving rhythmically, take selfies on command, and stabilize video automatically. When it goes on sale later this year, it will include an embedded AI agent developed in partnership with Alibaba one that runs using multiple AI models to handle complex, multi-step tasks.
This is a different vision of what a pocket size AI app device can be. Instead of just a tool, it is positioned as a companion something closer to the AI concepts people have seen in films and imagined for years. Whether users are ready to form that kind of relationship with their phone remains to be seen, but Honor is clearly betting they will be.
StepFun’s STEPX Neo: Built for Agents from Day One
Shanghai-based AI startup StepFun took a different approach entirely. Rather than adapting an existing phone to support AI agents, it built the STEPX Neo described as an “AI agent-native smartphone” from the ground up with agenting as the core function.
StepFun’s chairman Yin Qi announced deep partnerships with several major Chinese platforms, including Alipay and ride-hailing service Didi. Through these integrations, the STEPX Neo can handle travel bookings, everyday purchases, local service requests, office productivity tasks, and even video editing all through conversational commands.
This is the clearest example yet of what a true pocket size AI app experience looks like when the entire device is designed around it rather than built on top of a traditional smartphone architecture.
AI PMX: What Is Actually Happening Under the Hood
The term AI PMX referring to AI-powered mobile execution captures something important about this generation of devices. These are not phones that run a single AI app in the background. The AI is the operating system. It sits at the center of every interaction and acts as the user’s primary interface with the digital world.
In the context of the latest article on AI smartphones, what makes this moment different from previous AI phone announcements is execution depth. Earlier AI phones could transcribe your voice, suggest replies, or summarize articles. These new agentic phones can plan, execute, monitor, and adapt across multiple apps, in real time, without requiring the user to manage each step manually.
Marc Einstein of Counterpoint Research put it plainly: “There is no clear winner in this race yet, which is why it is currently quite a hot topic.” But he went further, predicting that within five to ten years, the way people use apps on their phones today will be unrecognizable. “This will fundamentally change the digital economy and disrupt business models,” he said.
Anwar AI and the Global Race
China is not the only player in this space, but it is moving the fastest. Outside China, global tech companies like Google are advancing AI smartphone capabilities steadily, with features like appointment booking already rolling out on Android devices.
US startup Brain Technologies launched what it calls the “Natural AI Phone,” which went on sale in Japan in April through a partnership with SoftBank. At a demonstration in April, the phone sent a message to a contact apologizing for being late based entirely on a single audio command. It was impressive, even though it still occasionally failed to complete tasks correctly.
This mirrors what the broader AI smartphone race looks like right now. The concept of Anwar AI a frictionless, always-available intelligent system embedded directly in your phone is clear and compelling. The technology is real and working. But it is not yet perfect, and the ecosystem battles between AI agents and app platforms are far from resolved.
What is increasingly clear is that the companies who figure out how to build an open, cooperative AI agent infrastructure one where apps and AI agents work together will define the next decade of mobile technology.
Global Impact
The implications of mainstream AI agent smartphones go well beyond convenience. They represent a fundamental shift in how digital economies operate.
Consider the advertising industry. If an AI agent is making purchasing decisions on a user’s behalf, traditional digital advertising which relies on capturing the user’s attention at key decision moments becomes far less effective. The AI agent becomes the decision-maker, and brands will need to find entirely new ways to reach consumers.
Consider the app economy. If a single AI agent can access and operate dozens of apps on a user’s behalf, many of those apps may become invisible to the user entirely. The front-end experience disappears, and what remains is just the outcome a delivered meal, a booked flight, a sent message.
China’s President Xi Jinping, speaking at the same conference, called for global cooperation in AI development and framed the technology as a shared human achievement rather than the property of any single nation. He announced that China would provide AI training opportunities to 50 developing countries and expand AI cooperation across multiple international organizations. Pakistan was among the 29 countries that signed an agreement at the conference to join a new World Artificial Intelligence Cooperation Organization headquartered in Shanghai.
What Experts Are Saying
Industry analysts are watching this space very closely. The consensus is that the technology is real, the demand is genuine, but the ecosystem complexity is significant.
Kiranjeet Kaur of IDC summarized the challenge: the AI agent model requires trust and cooperation from app developers, platform companies, and regulators simultaneously. Without that cooperation as the Doubao Phone incident showed even the best AI agent can be crippled overnight by a single API block.
Marc Einstein of Counterpoint Research offered a longer-term view. He believes the disruption is inevitable. The only question is timing. “In five or ten years time, we won’t be using apps on our phones like we do today,” he said.
Conclusion
The pocket size AI app phone is no longer a concept on a whiteboard. It is a product you can hold in your hand, talk to, and watch as it gets things done on your behalf. Whether it is Nubia’s NaviX Ultra, Honor’s Robot Phone, or StepFun’s STEPX Neo, the message from Shanghai this weekend was consistent: the age of the agentic smartphone has begun.
The road ahead is not without obstacles. App platforms will continue to push back. Technical performance will need to improve. Regulatory frameworks around AI agent behavior will need to evolve. But the direction is clear, and the momentum is real.The question is no longer whether AI will transform the phone in your pocket. It is how soon.
Frequently Asked Questions (FAQs)
What is the smallest AI computer?
When people ask about the smallest AI computer, they are usually referring to two different things: hardware-level AI chips and AI-powered consumer devices. On the hardware side, companies like Nvidia and Qualcomm have developed extremely compact AI processing chips some small enough to fit inside earbuds or wearable devices. But in practical terms, the most accessible and powerful small AI computers today are the latest AI-native smartphones. Devices like the Nubia NaviX Ultra, StepFun’s STEPX Neo, and the Honor Robot Phone pack enough AI processing power to run complex multi-step agent tasks entirely on-device, without needing to send every command to a cloud server. Qualcomm’s Snapdragon AI chips and Apple’s Neural Engine are examples of AI processors that bring serious computational intelligence into a phone-sized form factor. The race to shrink AI hardware continues, and 2026 is seeing some of the smallest yet most capable AI computing packages ever built for consumer use.
What is the size of AI?
This is a question with multiple answers depending on what you mean. If you are asking about the size of AI models measured in parameters today’s largest language models like GPT-4, Gemini Ultra, and China’s Doubao operate on hundreds of billions of parameters. These models require massive data center infrastructure to train and sometimes to run. However, the trend in 2026 is toward smaller, more efficient AI models that can run directly on devices like smartphones. These are called on-device or edge AI models, and they are designed to deliver strong performance with far fewer parameters sometimes as few as 1 to 7 billion while running entirely within the processing power of a pocket size AI app device. If you are asking about AI as an industry, it is enormous and growing rapidly. The global mobile AI market alone is estimated at between $25 billion and $32 billion in 2025 and is projected to grow to over $258 billion by 2034. AI, in other words, is getting smaller in its hardware footprint while simultaneously getting larger in its economic and social impact.
Who are the Big 4 of AI?
The Big 4 of AI is not a formally defined group, but the term is commonly used to refer to the four companies that have had the most dominant and foundational influence on the development of artificial intelligence as a global technology. In the US context, the Big 4 of AI are generally considered to be Google (Alphabet), Microsoft, Amazon, and Meta. Google pioneered the Transformer architecture that underpins nearly all modern large language models and runs Gemini. Microsoft made a transformative early investment in OpenAI and has integrated AI across its entire product ecosystem through products like Copilot. Amazon built the AI infrastructure backbone that powers much of the industry through AWS and developed its own models and Alexa ecosystem. Meta has become one of the most significant open-source AI contributors through its LLaMA model family, making powerful AI available to developers worldwide at no cost. Some analysts now also include Apple in this conversation given its deep AI integration across hundreds of millions of devices and OpenAI in its own right given the global dominance of ChatGPT. In the Chinese context, the equivalent group includes Baidu, Alibaba, Tencent, and Huawei often called China’s Big 4 of AI, with ByteDance increasingly seen as a fifth major player through tools like Doubao and its enormous data advantages.
