Source | Bohu Finance (bohuFN)

The Nubia NaviX Ultra recently hit the shelves, marking the debut of ByteDance's Doubao mobile AI assistant in its consumer version. Dubbed the "second-generation Doubao phone," it starts at RMB 5,999 (discounted to RMB 5,499 post-national subsidies).

Less than a year after the first-gen Doubao phone's launch, ByteDance and Nubia have rushed out a mass-produced successor, signaling a sense of urgency. Yet, in today's chaotic AI phone landscape, this move is somewhat understandable.

Over the past year, smartphones have emerged as a pivotal platform for on-device AI. Brands such as Honor, Vivo, and OPPO have made AI assistants a central selling point for their new models, while large model firms like JieYue XingChen are also keen to enter this market. Across the ocean, OpenAI is developing its own phone hardware.

Everyone is vying for an early foothold in this AI revolution. However, the question that piques the outside world's curiosity the most is how the second-gen Doubao phone will tackle the issue that plagued its predecessor: its ambition to "handle tasks for users" was thwarted by several mainstream apps.

ByteDance has yet to offer a flawless solution. Numerous bloggers have discovered through testing that the second-gen Doubao phone still faces restrictions in many commonly used apps. Recently, some netizens reported being abruptly logged out of Honor of Kings while using the second-gen Doubao phone, sparking heated discussions.

Despite introducing new rules, only a handful of apps are willing to open their ecosystems to the Doubao phone. As long as ecosystem providers wield veto power, outsiders will find it challenging to truly break in.

01 Still Confronting Ecosystem Barriers

Of course, the second-gen Doubao phone has also made strides.

So far, the second-gen Doubao phone has achieved impressive initial sales, with official reports indicating that it generated over RMB 100 million in sales within the first second of availability.

Evaluations from major media outlets and bloggers have highlighted several notable improvements in the second-gen Doubao phone.

In terms of design, the overall device is lighter and more portable. An AI button is conveniently placed on the side, enabling users to activate different AI functions with a single or double click. The operation is smoother, and security and privacy concerns can be addressed through fingerprint unlocking.

Regarding AI functionality, which is of paramount interest to users, the second-gen Doubao phone offers both surprises and disappointments.

On the positive side, the first-gen Doubao phone's standout feature—allowing users to invoke the AI assistant with a single sentence—has been further refined in the second generation, which not only performs tasks but does so more "intelligently."

In a video by Bilibili blogger "Xiaobai Testing," when a user says, "Help me book a ride to XX," the Doubao phone automatically executes the task, confirming the address and placing the order.

It can also assist users in booking flights and hotels across different apps, searching for music and books based on user instructions, and integrating more deeply with Feishu to summarize meeting recordings, generate PPTs and documents, effectively transforming the phone into an office assistant.

Moreover, the second-gen Doubao phone is more intelligent and proactive than its predecessor when performing these routine tasks.

For instance, after receiving a user's instruction, Doubao can decide whether to activate deep thinking mode. Its memory has also improved, allowing it to retrieve data from the user's phone and provide more personalized services...

Unfortunately, the second-gen Doubao phone still cannot operate freely across all apps.

In demonstrations of the first-gen Doubao phone, users could issue a command, and the phone would automatically compare prices in shopping apps and add items to the cart, with the user only needing to complete the final payment.

However, the second-gen Doubao phone can currently only perform these actions on designated shopping platforms. For food delivery, platforms like Taobao Snap Up, Meituan, and JD do not support order placement. Likewise, attempts to like posts on Xiaohongshu or WeChat Moments fail.

In short, many mainstream apps still impose restrictions on the Doubao phone. Netizens joke that it is essentially a ByteDance-exclusive device, as tasks like ordering food, watching videos, or listening to music must be performed through ByteDance's ecosystem apps.

The persistent "barrier" that the Doubao phone cannot overcome is the ecosystem walls between different apps.

The Doubao phone aims to become the sole entry point for users to handle tasks, but every app wants to keep users within its own ecosystem. This is not just a battle for entry points but also for profits.

The Doubao phone's approach has shifted from "forcing its way in" to "knocking on the door."

Two days before the launch event, the Doubao mobile assistant officially released the "Screen Automation Operation Statement Protocol (SAEP)," giving third-party apps the choice of whether to allow access.

Simply put, whether the Doubao phone can invoke an app depends on three steps:

First, it checks whether the app has adopted the MCP (Model Context Protocol) or A2A (Agent-to-Agent) protocol. If an agreement is in place, Doubao will send the request directly to the app without opening its interface, only providing feedback on the results.

Second, if no agreement exists, the Doubao phone will follow SAEP rules. The app can declare boundaries at three levels: page, business intent, and operation type. The Doubao phone will execute tasks only within the allowed scope.

Third, if the app explicitly rejects access from the Doubao phone, the Doubao phone will only help the user open the app, requiring manual task execution.

This time, the Doubao phone has adopted a humble stance, allowing third-party apps with ecosystem capabilities to decide how much and how deeply the Doubao phone can operate within a stricter compliance framework.

02 Each Player Has Its Own Strategy

However, by surrendering control, the second-gen Doubao phone can accomplish fewer tasks, and its most attractive selling point has become a promise that cannot be fully fulfilled.

For the Doubao phone, this was a reluctant choice.

From the first to the second generation, the Doubao phone's goal remains unchanged: to become the user's AI assistant, operating the entire phone on behalf of the user, transforming it from a manually operated tool into a super butler that understands and accomplishes tasks.

This vision is appealing, but too many agents, platforms, and smart hardware devices are vying for this position. Each player has its own business considerations, and none will fully open its ecosystem to competitors without calculating the costs.

However, business deals can still be struck.

Apps want to protect their entry points, but they also covet ByteDance's massive traffic. Therefore, while they may block the first-gen Doubao phone, which directly intrudes without permission, they are willing to negotiate with the second generation.

Current tests show that the Doubao phone still cannot operate directly on mainstream apps like Taobao, Alipay, Meituan, WeChat, and Xiaohongshu. However, SAEP leaves room for future integration with these apps.

Whether the Doubao phone can ultimately join these ecosystems and access their functions depends on the mutual cooperation sincerity.

QuestMobile data shows that by June 2026, Doubao had 382 million monthly active users. ByteDance's traffic could bring new growth opportunities for transaction-based apps.

However, as ByteDance gains more user profiles and shopping preferences, its influence over user transaction flows will grow. For ByteDance's direct competitors, the calculations may not be so straightforward.

Interestingly, WeChat has already opened A2A capabilities to Huawei, Honor, Xiaomi, OPPO, and Vivo this year but has excluded Nubia and Doubao from its cooperation list.

Clearly, the Doubao phone is not giving up on the AI phone narrative.

In fact, to "handle tasks for users," the Doubao phone does not necessarily need a dedicated device. Currently, Doubao's transaction module has gradually integrated into e-commerce, local services, and travel sectors. Users can also "accomplish tasks with a single sentence" through the Doubao app.

However, this model has its limitations. No matter how capable Doubao is, it operates on others' hardware. Issues like how much data Doubao can access or how deeply it can integrate with system interfaces are not under its control.

Therefore, Doubao must create its own "body." With its own hardware, it can gain full administrative privileges, completely read user data, understand context, and have the opportunity to define the next generation of smart interactions.

However, this position is not exclusive to the Doubao phone. Other phone manufacturers, smart hardware companies, and large model enterprises like Alibaba are also competing for this "entry point" through collaborations with hardware vendors.

Phone manufacturers guard the hardware entry points, while third-party apps protect their ecosystem entry points. AI phones and agents hoping to advance and handle tasks on behalf of users must confront these two barriers.

They can either build everything from scratch—systems, ecosystems, and AI capabilities—or negotiate cooperation, ceding some control in exchange for entry.

03 Who Will Dominate the Phone?

Of course, this is not a problem unique to the Doubao phone. All agents seeking to control entry points face the same two barriers.

This is why cross-border collaborations between large model companies and phone manufacturers have intensified over the past six months, as neither can overcome these challenges alone.

However, new questions arise. If agents cannot break through the app ecosystem barriers, consumers will struggle to obtain an AI assistant that can "handle everything with a single sentence."

Is the AI phone still the best form for agent deployment?

A reference comes from overseas. In September this year, Meta launched its personal AI assistant, Muse. Within a week, it topped the U.S. App Store's free app chart.

Its popularity stems not from groundbreaking technology or experience but from addressing the common issues of previous cloud-based agents: while the "lobster" (a metaphor for agents) could do everything, it did nothing exceptionally well.

Therefore, Muse was designed as a standalone app, eliminating installation barriers. It uses independent virtual machines to address security and memory issues, offers free tokens to users, and primarily relies on transaction commissions for revenue.

Most importantly, Muse has identified scenarios that truly retain users, focusing on tasks like email handling, form filling, and travel bookings for the general public. Its popularity proves the immense potential demand for consumer-grade agents.

However, domestic agents cannot simply replicate Muse's path. Not because domestic agents are less intelligent, but because control does not lie with them.

Overseas, email and credit card payments are infrastructure, and web pages lack complex interest conflicts, allowing Muse to complete tasks with minimal friction.

In China, mobile internet services are consolidated within a few super apps. When agents attempt to perform tasks for users, they first encounter "ecosystem walls," making the process more convoluted.

However, optimistically, while the current fragmented landscape of domestic apps is chaotic, it also means each ecosystem can independently complete a closed loop of "social + content + functionality + payment." Once ecosystem interfaces are opened, "phones handling tasks for users" could become a reality.

However, taking this first step will not be easy.

The Doubao phone is well aware of this. According to media reports, the second-gen Doubao phone has an overall inventory of about 200,000 units, with less than 100,000 units in the initial batch. From an expected shipment perspective, the Doubao phone remains a "test field."

Even so, other phone manufacturers and large model companies cannot afford to be complacent. The Doubao phone enters the market with unresolved issues, aiming to address them through iterative product improvements.

Ultimately, the biggest challenge facing AI phones today is not technical but ecological.

The number of ecosystems the Doubao phone can integrate depends on the leverage ByteDance can offer. With sufficient patience and sincerity, the current interests landscape could be reshaped.

Entry points are the facade; rooting is the foundation. Only by truly infiltrating users' daily lives and work can agents gain firmer control than mere entry points.

Therefore, by entering the market during a window when other phone manufacturers are intensively launching new products, the Doubao phone at least provides consumers with an additional choice and reference. Regarding which AI is better, the Doubao phone offers an answer worthy of examination.

The cover image and illustrations belong to their respective copyright holders. If the copyright owners believe their works are unsuitable for public browsing or free use, please contact us promptly, and our platform will make immediate corrections.