Doubao phone has stirred up a tremendous wave.

```

Author | Huang Yu

Editor | Zhang Xiaoling

Eighteen years ago, Apple ushered in a smart era of phones built around independent apps with the launch of the iPhone, establishing the "walled garden" ecosystem that remains today. Eighteen years later, another signal of software ecosystem reform is sounding, but this time the firestarter is not a hardware giant, but a new generation of internet giants looking to rewrite the software order.

On December 1, ByteDance and ZTE jointly launched the “Doubao Phone”—nubia M153—making waves throughout the tech industry. This phone, which deeply embeds the Doubao assistant in its system, saw its initial batch of 30,000 units sold out instantly upon release.

Meanwhile, Doubao has provoked the most sensitive nerves in the mobile internet ecosystem.

The Doubao Phone Assistant is an “AI commander” with system-level permissions, able to simulate human actions—directly challenging the traditional role of APP developers as "data controllers." This has triggered fierce debates about data authorization, privacy infringement, and system security, while shaking the foundations of the super APP commercial ecosystem and traffic base.

Currently, WeChat, Taobao, and certain banking apps have set up “security firewalls” against Doubao Phone Assistant, starting to counter and block its operations.

ByteDance, which has pushed Doubao into the top spot for AI-native applications in China, has showcased the ability of AI Agents to operate across apps via this "engineering prototype," declaring a full-scale upgrade in the battle for the next-generation super entry point.

ByteDance’s offensive and Tencent’s defense resemble a rerun of the past Toutiao-Tencent wars: from the mobile internet era to the AI era, tech giants have never stopped competing for the super traffic gateway.

This is not just a contest of technical paths, but a multi-party game involving phone manufacturers, internet giants, and AI startups. In this battle to reshape the future software ecosystem order, no party can detach itself.

The Hand of God

On traditional smartphones, each APP is like an isolated island, with data walled off to protect user privacy and operation autonomy. Over the long history of phone development, this sandbox isolation mechanism gradually formed.

However, with this collaborative phone, Doubao Phone Assistant acts as an “AI commander,” capable of invoking services across apps with a single spoken command: ordering takeout, booking flights, comparing shopping prices, responding to WeChat messages, playing mini games, and more.

Doubao achieves cross-APP automation via the system-level permission INJECT_EVENTS—regarded as the “Hand of God” in the Android ecosystem—which simulates user clicks and reads screen content. Its core power is breaking down app barriers imposed during the mobile internet era.

This embodies the revolutionary attempt by AI Agents to replace traditional apps as users’ task managers. Yet, the ability to operate across apps quickly met resistance from established ecosystem players.

As a national-level app, WeChat was among the first recognized by some Doubao users as refusing authorization to Doubao Phone Assistant.

On the night of December 2, multiple users reported that while using Doubao Assistant on nubia M153 to operate phone functions, any operation involving WeChat led to abnormal WeChat exits or even inability to log in.

WeChat personnel told Wallstreet News: "Nothing unusual was done, perhaps it triggered existing security risk controls."

Wallstreet News also learned that Taobao began to prompt frequent human-machine verification or experienced forced logouts and app crashes when used with Doubao Assistant.

Meanwhile, banking apps such as CCB and CMB now prompt users that the app cannot be used when recording the screen.

This series of issues exposes the legal and compliance challenges faced by AI Agents in implementation.

Doubao has responded to these problems.

On December 3, the Doubao team announced that, after receiving user feedback, the capability to operate WeChat via Doubao Assistant has been withdrawn. Phones previously banned from logging in to WeChat on the nubia M153 are being gradually unblocked, and users can wait and try to log in again.

Doubao reiterated that all industry AI assistants must use the INJECT_EVENTS (or similar accessibility permissions) for cross-app auto-operation, and user authorization is required; Doubao discloses this in its permission list.

This is the crux of the dispute between Doubao and various APPs: Does user authorization replace third-party platform authorization? WeChat, Taobao, and others have obviously not approved it.

To defuse the turbulence, Doubao stated it places great importance on user privacy: for example, screens and operations are never stored on cloud servers, nor entered into model training—ensuring user privacy and security.

Doubao also stressed that the current release is a “technical preview,” not for regular consumers, but aimed at industry and AI technology enthusiasts. Some features and interactions are innovative and currently lack clear industry consensus.

With conflicting interests and data security concerns, some third-party APPs are clearly reluctant to open authorization to AI Agents. Thus, Doubao is taking a step-by-step approach, opting to launch as an “engineering prototype.”

The Weight of Security

Doubao Phone’s “Hand of God” appears to open Pandora’s box; in fact, Doubao is not the first to launch an AI phone.

2024 is dubbed “the first year for AI phones,” with Huawei, Honor, OPPO, vivo, and others all releasing AI phones, integrating AI into the system and touting the slogan “let your AI phone assistant do the job with a single sentence.”

Yet, most attempts to date are mere icing on the cake—feature optimizations reliant on broader app authorization, more akin to mobile internet era “local improvements,” without touching the foundational APP ecosystem nor driving much user excitement.

When risk controls are touched, WeChat immediately counters specific AI Agent permissions. In fact, WeChat's boycott of AI Agents is nothing new—it warned about third-party chat log theft in April and closed off a batch of AI permissions in August, even disabling Huawei’s Xiaoyi and Xiaomi’s Xiaoai from accessing WeChat.

For an AI phone to truly become an AI Agent, two technical paths exist: One, the intent framework, requires third-party app authorization, is mature, and has low hardware demands; the other, a pure visual scheme, needs no authorization but requires higher hardware capabilities and further technical development.

So far, phone makers are more cautious with cross-app operations for AI Agents, generally preferring the intent framework model.

For example, vivo released an intent framework white paper last year; this year, intent framework upgraded to 2.0, fully compatible with MCP protocol.

Meanwhile, vivo introduced the smart agent protocol A2A. Vivo Vice President, OS Product VP, and AI Global Research Academy Director Zhou Wei said: “With this, developers can create agents and configure cards entirely online via the agent protocol, distributing them across all vivo channels through the unified intent framework and agent container.”

Negotiating agreements with more third-party APPs is undeniably hard.

To reduce resistance, ByteDance took a step back. On December 5, Doubao announced plans for further normalization and restrictions of AI phone operation capabilities, especially on usage scenarios like points farming and incentives, financial apps, etc.

Meanwhile, ByteDance insiders said they are actively seeking deeper discussions with app vendors to form clearer and more predictable rules, avoiding blanket prohibition of users' legitimate use of AI.

In summary, in the battle over AI era discourse power, phone makers and Doubao face clear development bottlenecks in exploring system-level AI powers due to authorization difficulties.

Wu Shenkuo, PhD supervisor at Beijing Normal University's School of Law, said AI Agents are innovative efficiency tools and a technological harbinger for the next stage of business model and ecosystem iteration. In governance, multi-party deep insights and judgment are needed, balancing tech innovation with industry interests.

AI Agents must globally access and understand user data and services from various APPs—once a system-side large model gains full data access, guaranteeing that user privacy is not leaked and system security is not breached becomes the "Sword of Damocles" hanging over all stakeholders.

This contest over authorization and boundary-crossing is far from over.

The Gateway War

The reason Doubao Phone Assistant is at the center of attention, besides privacy concerns, is its unprecedented challenge to the mobile internet industry landscape of the past decade. The core of this war is the battle for the “super traffic gateway” of the AI age.

Since Apple pioneered smartphones, a super commercial ecosystem spanning tens of billions of phones and countless APPs has formed—comprising hardware, OS, super apps, and of course the underlying network and telecoms.

Internet and tech giants built on national-level super apps have also emerged. In China, WeChat, Douyin (TikTok), Pinduoduo, Xiaohongshu are standouts; Taobao, JD.com, Meituan and other traditional e-commerce and services sites have also grown into super apps.

The core of the super app business model is data and traffic. On this foundation, mature models of advertising, commerce, and payment have developed—a super app may involve trillions in GMV, hundreds of billions in revenue, and tens of billions in profit.

Doubao aspires to become the “traffic dispatch center” via AI assistant status, bypassing traditional app entry, and invoking services directly. This disrupts current APP distribution and usage, hitting at the heart of the super app business model.

For platforms like Taobao and Meituan, which rely on user time, ad exposure, and transaction commissions, AI Agent’s cross-app scheduling means users need not open the APP to consume services—directly siphoning away the platform’s traffic foundation.

A gateway is both a moat and a toll station. This new-old power struggle ignited by Doubao is essentially the lords of mobile internet era traffic fiefdoms countering emerging AI Agent’s “cash register pure takeover” attempts.

Since the end of 2022 when ChatGPT set off an AI large model wave, there's a consensus that AI Agent will bring unprecedented change to the software and hardware industries.

Tesla CEO Musk recently predicted that in 5-6 years, traditional phones and apps will disappear, with most human-consumed content generated by AI.

While reshaping human-computer interaction, AI Agent will also change the APP developer’s data controller role and rewrite the mobile app business model—after "Apple Tax" and "Android Tax," new forms of business relationships will follow.

Thus, the war started by Doubao is a battle between the established "Super App Empires" and emerging "AI Agent Models," fundamentally a fight for the future super traffic gateway.

ByteDance faces traffic anxieties, and its proactive offensive met inevitable super app risk control barriers. This is not WeChat or Taobao versus Doubao per se, but a clash of old and new models.

The Battle for the Future

Zhang Yiming is still the man determined to smash the old world. Over a decade ago, he used Douyin (TikTok) to upend social media, creating a national-level app to rival WeChat; now, he wants to use Doubao Assistant to seize the super traffic gateway of the AI era.

Doubao is actively attacking; WeChat and Taobao appear to be passively defending, yet as mobile internet era overlords, Tencent's Ma Huateng, Alibaba's Jack Ma, even Baidu's Robin Li are not to be outdone, and all have their own AI Agent ambitions.

Tencent’s management revealed early this year that in general AI Agent capability, the company is building with native AI products like Yuanbao and IMA, and will also develop agents integrated with and running inside the unique WeChat ecosystem.

Tencent President Liu Chiping said in a recent quarterly call that WeChat will eventually launch an AI agent to help users complete many tasks within WeChat. With communication, social, content, mini-program, payment and business ecosystems, WeChat can help intelligent agents understand user needs, intentions, and interests, and close the loop on execution.

Alibaba has also outlined strategic plans for its own AI Agent. Its "Qianwen" APP is the vanguard for grabbing next-gen super traffic gate, with the core management tabbing it as "the future battle of the AI era." Qianwen will integrate with e-commerce, maps, local life, and other Alibaba business scenarios.

Beyond the internet giants, phone manufacturers with underlying OS access are also contending to let AI take over phones. The AI assistant is a critical landing scenario for AI phones; most makers build their own: Huawei Xiaoyi, Honor YOYO, Xiaomi Xiaoai, OPPO Xiaobu, etc.

This time, ByteDance partnered with ZTE on an AI phone in part because ZTE lacks its own AI assistant. This is similar to Huawei’s Harmony Intelligence originally only cooperating with Seres—most car brands are unwilling to “give their soul” to Huawei, and phone makers are likewise cautious.

Indeed, besides ZTE, ByteDance has its eyes on Smartisan phones. Alibaba already has deep cooperation with Apple, and may even launch its own phone.

In addition to phones, smart hardware includes glasses, car-mounted screens, humanoid robots, etc.—all of which need to balance with AI assistants.

In future, AI app and hardware vendor cooperation will be deeper and more frequent; makers may jump into hardware production as Alibaba has done with its Quark glasses. Without install base, without users, AI assistants can’t gain traction.

We can imagine that in the foreseeable future, several internet giants will have AI Agents connecting their respective ecosystems, even making hardware themselves; phone makers may work with more small/medium APPs to build more capable system-level AI Agents.

A layered ecosystem seems unavoidable: highly sensitive services (payments, social) will be tightly controlled by platforms; low-sensitivity cases (retouching, info queries) may be open to AI Agent scheduling.

Doubao’s attack and WeChat/Taobao’s countermeasures mark the transition of AI phones from simple tech demos to a brutal “ecosystem battle.” Doubao’s temporary setback exposes the limits of technological radicalism against entrenched interests.

But this collision is not bad—it will accelerate industry reflection and reconstruction of future human-computer interaction rules. The terminal AI era will surely disrupt the traditional mobile internet privacy and data security order, forcing all sides to find new balances; the old smart terminal and super APP landscapes will be restructured.

It is indisputable: the revolutionary significance of AI Agents is already established, and they are irreversibly reshaping the future appearance of software and hardware industries, ushering humans and AI/interacting agents into a new era.

The future is here. ByteDance and the Doubao phone are simply the first to lift the veil.

Risk Warning and DisclaimerMarkets carry risks, investment requires caution. This article does not constitute individual investment advice, nor does it consider the particular investment objectives, financial situation or needs of any one user. Readers should consider whether any opinions, views, or conclusions herein fit their individual circumstances. Acting on this article is at your own risk. ```