A well-rounded AI smartphone should exercise a bit more 'self-control'.

Recently, there's been a buzz about the second-generation Doubao AI smartphone. Industry insider Xinliu Think Tank exclusively reported that ZTE Nubia has shifted its focus entirely, putting all its core resources into this new AI smartphone, with plans to unveil it within the month.

The countdown has begun.

Although ZTE, Nubia, or ByteDance haven't officially confirmed this yet, public information from the past six months indicates that Nubia and ByteDance have been pouring resources into developing the next generation of Doubao AI smartphones. This includes Nubia President Ni Fei's sneak peek of this 'new smartphone breed' at MWC in late February this year.

Image Source: Weibo

Three to four months ago, the first-gen Doubao smartphone made its debut as the Nubia M153. Despite being a 'prototype' with a technical preview of the Doubao smartphone assistant, the product was already impressively refined.

What really caught everyone's attention—and sparked a lively discussion—was its ability to take natural language instructions and perform cross-app tasks like price comparisons, photo editing, ticket searches, order placements, and messaging. In some cases, it could even mimic a real person using a smartphone, opening apps step-by-step, recognizing interfaces, and completing tasks.

The first-gen Doubao smartphone took the concept of 'AI smartphones,' which manufacturers had been discussing for years, and brought it to the practical stage of 'AI doing the smartphone work for you.'

However, when AI starts doing more than just answering questions—clicking screens, launching apps, accessing galleries, handling payments, and managing social connections via GUI (Graphical User Interface)—it inevitably runs into permission and privacy issues, disrupting today's internet business ecosystem.

The first-gen Doubao smartphone quickly hit this snag. WeChat, Alipay, banks, shopping platforms... all instinctively tightened their restrictions on its invocations and operations, sparking broader discussions about system-level permissions, simulated inputs, account security, and privacy boundaries.

This is the challenge the second-gen Doubao AI smartphone must tackle: It can't just be faster, pricier, or more flagship-like than its predecessor—it must also overcome privacy hurdles, evolving from a 'prototype' into a mass-market device that everyday users can trust.

Hardware Upgrades to Accommodate Agents

Right now, details about the second-gen Doubao AI smartphone's hardware are scarce. The most concrete news is that it's expected to be powered by the fifth-gen Snapdragon 8 Elite. Considering the first-gen M153 already sported the Snapdragon 8 Elite, 16GB+512GB storage, a 6.78-inch LTPO screen, and a 6000mAh battery, it's no surprise that the second gen continues with a flagship platform.

First-Gen Doubao AI Smartphone. Image Source: Lei Technology

From a traditional smartphone standpoint, these specs aren't groundbreaking. By 2026, which Android flagship doesn't boast a top-tier chip or a hefty battery? But the real evolution of the second-gen Doubao AI smartphone lies in redesigning hardware around Agents.

Historically, smartphone hardware has prioritized Apps. Chips ensured speedy app launches, screens guaranteed excellent display quality, imaging systems delivered strong photography capabilities, and batteries lasted a full day. With the addition of AI Agents, smartphones now handle a new category of persistent tasks:

They must interpret user instructions, recognize screen content, invoke cameras, microphones, location services, galleries, calendars, notifications, and app states at any time, make decisions between cloud and on-device models, and do so without significantly slowing down the system or increasing heat and power consumption.

This means the second-gen Doubao AI smartphone needs more than just a beefier SoC—it requires a holistic system engineering approach centered around on-device AI.

Additionally, according to Qualcomm's description of this platform, beyond continued improvements in CPU, GPU, and NPU performance, the fifth-gen Snapdragon 8 Elite's core upgrades include on-device learning, real-time perception, personal knowledge graphs, and Agentic AI capabilities.

If the second-gen Doubao AI smartphone adopts this chip, it should primarily leverage its on-device capabilities. For example, it could process a portion of personal memories, preferences, frequent contacts, and common task workflows locally. When a user says, 'Book me a ticket to Guangzhou tomorrow,' it shouldn't start from scratch each time asking for preferences—instead, it should know the user's preferred seat class, travel app, invoice details, and whether they prefer morning departures.

Image Source: Qualcomm

The more comprehensive the on-device memory, the more the AI resembles an assistant that truly understands user habits.

Multimodal understanding should also rely more on on-device processing. When users ask questions like 'Is this reliable?' or 'Summarize this' or 'Send this address to him' from any interface, the AI needs to quickly interpret screen content. Uploading screenshots to the cloud each time strains speed, privacy, and stability.

Stronger NPUs, memory, and local models can enable these lightweight tasks to be completed directly on the smartphone.

Another often-overlooked aspect: thermal management and battery life. Traditional flagship smartphones face high loads primarily from gaming and imaging, which users can perceive and typically have defined durations. However, Agents may introduce more fragmented, frequent high loads. They may not always run at full performance but could remain active in the background all day, waiting, listening, recognizing, summarizing, and retrieving information.

Thus, the second-gen product will likely continue to feature a large battery and may see improvements in cooling, memory, storage, and system scheduling. Further speculation suggests its hardware design will emphasize several AI entry points: a dedicated AI button, higher-quality microphones, more stable voice wake-up, stronger screen content recognition, better privacy prompts, and a form factor better suited for prolonged holding and voice interaction.

From First to Second Generation: From 'Operation' to 'Teamwork'

Even more critical is the AI itself. Today, it's almost certain that the second-gen Doubao AI smartphone will undergo significant changes in its 'agent' approach, as the external environment has fundamentally shifted.

Image Source: OpenClaw

Over the past six months, heavyweight products like OpenClaw, Claude Code, and Codex have driven a crucial shift in the Agent ecosystem: internet platforms are accelerating their embrace of Agents, enabling interactions through protocols like MCP and A2A or official Skills.

MCP addresses how AI connects to tools and data sources. It transforms customized interfaces into a more universal connection method. For developers, AI no longer needs to write separate invocation logic for each service; for service providers, it allows them to expose their capabilities more standardly.

A2A, meanwhile, resolves how agents communicate with each other. The smartphone's system assistant can be an Agent, while WeChat, Alipay, Feishu, and Taobao can each have their own Agents.

The system assistant doesn't necessarily need to tap WeChat's interface like a human—instead, it can send a clear request to WeChat's Agent: send a message to a contact or initiate a video call. WeChat then executes this within its security boundaries and returns the result to the phone's assistant.

While this may sound like a mere technical shift, it's highly significant for AI smartphones. The first-gen Doubao smartphone attempted to 'operate Apps on behalf of users,' but the GUI-based Agent approach proved too disruptive to the existing ecosystem. In contrast, the protocol-based Agent approach is gaining traction.

WeChat's recent progress with multiple smartphone manufacturers on A2A assistant capabilities is a clear signal. WeChat hasn't fully opened its ecosystem, but it now allows phone system assistants to invoke WeChat capabilities in specific scenarios, such as sending messages or initiating audio/video calls. The entire process emphasizes dual authorization and execution by WeChat itself, which returns the results.

Image Source: Weibo

Doubao, too, has followed suit over the past six months, connecting its own e-commerce and payment services while also integrating third-party platform services. For instance, the Doubao app recently launched a gray-scale test for one-click ride-hailing in Beijing and Hangzhou, powered by Cao Cao Mobility. Users can simply state their travel needs in the chatbox, and the system automatically identifies the location, number of passengers, and preferences, matches routes and prices, and confirms the order with one tap.

Image Source: Weibo

Thus, it's foreseeable that the second-gen Doubao AI smartphone may retain GUI Agents, as countless mid-to-long-tail Apps cannot immediately adopt standard protocols. However, for high-risk services and dominant platforms, it will need more protocol-based, authorized connections.

Where A2A or similar mechanisms can be used, forced simulated taps should be avoided. When simulation is necessary, clearer permission prompts, operation replays, critical step confirmations, and risk interceptions should be implemented. This will make the second-gen Doubao AI smartphone seem less 'wild' than the first gen but closer to a device truly sellable to ordinary users.

A Mature AI Smartphone Should Be More 'Self-Controlled'

Over the past two years, the smartphone industry has talked extensively about AI, with many features sounding impressive but delivering minimal real change for users. The Doubao smartphone has jolted the industry, accelerating competition into the deeper waters of app ecosystems and operational permissions:

Smartphone manufacturers are redefining system assistants, internet platforms are redefining open boundaries, chipmakers continue to provide more powerful compute and energy efficiency for on-device Agents, and developers must consider how their Apps can be invoked, understood, and distributed by AI.

So, will the second-gen Doubao AI smartphone look like this? We cannot yet confirm.

However, a truly mature AI smartphone should exercise greater restraint in interactions between humans and Agents, as well as between Agents and devices: In most scenarios, it should minimize user operations, but in critical scenarios, it must clearly show users what the AI is doing. It can help users fill forms, compare prices, organize itineraries, edit photos, summarize files, and initiate communications—but for sensitive operations like payments, messaging, account logins, and financial transactions, there should be clear confirmations and traceable records.

On the other hand, as Lei Technology previously argued, AI smartphones should not treat GUI Agents as the only solution, nor should they entirely abandon the versatility advantages of GUI Agents. After all, for countless mid-to-long-tail Apps, developers lack the resources or incentive to adapt to Agent-based interactions immediately.

Similarly, AI smartphones cannot rely solely on cloud models—improvements in on-device AI capabilities are essential. Only through on-device low latency, minimal disruption, preference retention, and contextual understanding can daily experiences be ensured.

If the second-gen Doubao AI smartphone achieves these goals, its significance will extend beyond Doubao and Nubia.

Source: Lei Technology

Images in this article are from 123RF's licensed library. Source: Lei Technology