On October 14, Doubao Mobile Assistant's official WeChat account announced that the 30-day public comment period for the Screen Automation Execution Protocol (SAEP) had closed, and that the platform would gradually open screen automation permissions to third-party applications.

According to the public notice, the latest list of applications refusing operations includes Pinduoduo, Kimi, Kuaishou's suite of apps, Meituan's suite of apps, Xiaohongshu, DeepSeek, and Industrial and Commercial Bank of China, along with 23 Alibaba and affiliated ecosystem apps such as Taobao, Qianwen, and Alipay, and 65 Tencent and affiliated ecosystem apps such as WeChat, QQ, and Honor of Kings—151 in total.

As previously explained, SAEP is a set of AI operation rules for third-party apps. Developers can independently declare whether to allow AI assistants to perform screen automation operations on their own applications, defining operation boundaries and restriction scopes. Officials emphasized that after the public comment period ends, application developers can still submit refusal declarations at any time, and once received, Doubao Mobile Assistant will promptly disable the relevant operations.

01 The Foundation of Super Apps Lies Hidden in the Traffic Closed Loop

For more than two years, computing power anxiety was the core narrative of the mobile AI Agent industry—but it dissipated faster than many expected.

In just two years, the parameter scale of on-device large models has leaped from billions to tens of billions, and the NPU computing power density of flagship chips has increased significantly. Underlying engineering challenges—memory bandwidth bottlenecks, quantization compression algorithms, and on-device inference frameworks—have continued to advance through joint efforts by chip manufacturers and phone makers.

Today, many flagship models can already support basic AI capabilities locally, such as voice interaction, intent understanding, and image recognition. Multimodal models' perception of phone screens is no longer limited to recognizing button coordinates; they have begun to possess the ability to parse shallow page semantics.

Technological iteration is still accelerating, but a core contradiction previously underestimated by the industry is rapidly emerging: now that AI has gained the ability to understand screens, simulate clicks, and decompose and execute instructions across applications, what determines the boundary of mobile AI Agents is no longer the technical ceiling, but the access red line of the application ecosystem.

Computing power defines the capability ceiling of AI Agents, while ecosystem rules define their actual activity boundary.

The two are not on the same dimension at all: computing power bottlenecks rely on capital investment and technological iteration and have room for continuous improvement; but the reconstruction of ecosystem order leverages the commercial interest foundation that the mobile Internet has accumulated over more than a decade.

To understand the motivation behind this collective closure of permissions, it is necessary to return to the core business model that has operated in the mobile Internet for more than a decade: the traffic closed loop.

Over the past decade or so, the value chain of mobile Internet platforms has always operated around this closed-loop system: platforms use content, social relationships, product supply, and local services as anchors to retain user time and attention within their own app ecosystems, and then monetize value through information feed advertising, search bidding, transaction commissions, membership subscriptions, and other methods.

Every browse, click, price comparison, and order placement by users within an app is a carrier of traffic value; advertisers pay for exposure, merchants pay for traffic, users pay for services, and platforms extract revenue from every layer of value exchange.

The reason this traffic closed loop is sufficiently stable lies not in technical barriers, but in the fact that it is a business model that has been repeatedly verified and is self-consistent.

Search advertising is settled by clicks, information feed advertising measures value by metrics such as dwell time, e-commerce platforms take commissions based on transaction volume, and local life services share revenue based on in-store redemption—the premise for all monetization links is that users complete the full interaction chain within the app.

Platforms' control over user interaction paths is essentially the foundation of their commercial pricing power. Whoever controls users' browsing paths, decision-making processes, and transaction entrances possesses the dominant power in distributing the entire value chain.

Technology can be caught up with, and product features can be imitated, but the traffic distribution power accumulated from more than a decade of user habits constitutes the commercial foundation of the mobile Internet. Any technology that attempts to shake the premise that users complete full-link actions within apps may impact the foundation of the entire value chain.

The emergence of the SAEP protocol is a consultation mechanism built by phone manufacturers in response to this commercial reality.

It no longer unilaterally decides screen reading and automation permissions by phone manufacturers, but establishes an ex ante declaration mechanism: returning to developers the decision-making power on whether to allow AI assistants to perform screen automation operations on their own applications. Developers can independently define permission boundaries and set operation restrictions, and even after the public comment period ends, they can submit refusal declarations at any time.

The design idea of this mechanism is similar to robots.txt on the PC Internet, where developers independently define accessible boundaries and then explore possibilities for cooperation. But the premise for the SAEP mechanism to be implemented and effective is that developers are willing to cooperate.

Judging from the results of the first public notice, the degree of cooperation from leading platforms is far lower than industry expectations. Behind this collective refusal, it is far from being simply targeted at a certain phone manufacturer: super apps in different tracks all base themselves on their own commercial foundations, weigh the costs and risks of opening permissions, have their own commercial considerations, and their refusals all have corresponding realistic reasons.

02 Shift of Decision-Making Entrance, Platforms Face a Role Change from Front Office to Back Office

The core concern of apps is not distrust of AI technology itself, but the lack of compensation mechanisms for commercial losses caused by opening permissions.

The core demands of social and gaming platforms are to protect user time and immersive interaction, while AI proxy operations reduce users' willingness to actively open, browse, and interact with applications. User time, on which information feed advertising depends for survival, faces the risk of diversion. For social and gaming systems that rely on user time and advertising monetization, this may impact their commercial fundamentals.

The focus of e-commerce and local life platforms is to protect transaction links and traffic distribution power. If AI agents intervene deeply, cross-platform price comparison and automatic ordering are expected to become a potential trend, and the search ranking, advertising placements, and store traffic distribution mechanisms carefully designed by platforms may be bypassed in AI execution links.

If they lose control over user decision paths, the foundation of platform advertising and commission revenue faces the risk of being shaken.

The concern of large model manufacturers is essentially mutual fortification among similar products. AI products naturally compete directly. If competitor agents are allowed to automatically operate within their own applications, user behavior and conversation data may be indirectly collected by competitors, and product capability boundaries also face the risk of being probed.

In the large model track where data is a core asset, such open behavior is akin to exposing one's own cards to competitors.

The concern of financial apps stems from hard constraints at the levels of risk control and compliance. In screen automation scenarios, transaction responsibility allocation, abnormal transaction identification, and corresponding regulatory details still lack clear standards. For financial institutions, fund security and compliance have far higher priority than experience and efficiency improvement. Before relevant rules are clarified, closing permissions is the lowest-risk choice.

It is not difficult to see that the root cause of refusing openness is not that AI technology itself is unreliable, but that platforms anticipate possible damage to core interests after openness and lack corresponding compensation mechanisms. Social platforms worry about loss of time, e-commerce worries about transaction links being bypassed, large model manufacturers fear data leakage, and financial institutions prioritize preventing security risks.

The real impact of GUI automation AI Agents on the existing ecosystem is not that AI clicks screens on behalf of users, but that it may fundamentally facilitate the migration of user decision-making entrances.

The ideal scenario depicted by GUI automation is: users give a natural language instruction, and AI completes a whole set of task processes such as price comparison, ordering, reservations, and inquiries across applications. Users do not need to open apps one by one, nor will they browse homepage recommended content. Native advertising may lose touch opportunities, and the search traffic distribution mechanism painstakingly built by platforms may be weakened in AI execution links.

For platforms, this means a role change: from a business front office that directly receives users to a back office service invoked by AI. The front office controls traffic, advertising placements, and pricing power. After becoming the back office, it can only passively respond to invocation requests, and how to realize revenue is a difficult problem many platforms need to face directly.

What is more worth vigilance is that once such user habits are formed, the cost of subsequent reversal will rise significantly.

The logic of platform choice thus becomes clear: rather than waiting for AI agents to gradually divert user decision-making links, it is better to define boundaries in advance. Refusing GUI automation is not resistance to technological progress, but protection of their most core asset—the dominant power over user interaction.

Computing power has room for continuous optimization and upgrading, but the user habits, traffic distribution power, and transaction links that super apps have accumulated over more than a decade will not automatically reconstruct with technological iteration.

It may even form a reverse game: the stronger the Agent capabilities and the deeper the penetration into scenarios, the stronger the platforms' vigilance, and the more the ecosystem barriers may continue to solidify.

03 The Second Half of AI Agents: The Technology Race Ends, the Interest Game Begins

The narrative of all-powerful Agents once highly expected by the industry is being continuously revised in the face of ecosystem reality.

The first route is GUI screen automation, namely the solution corresponding to the SAEP protocol. Its advantages are intuitive: no need for apps to separately develop interfaces, low implementation threshold, and theoretically coverage of a large number of application scenarios.

But its shortcomings are equally prominent. This capability is highly dependent on the willingness of application parties to cooperate, and a single statement from the platform can cut off the function; page revisions may cause recognition failure, and a series of issues such as security audits and responsibility allocation remain undecided to this day.

This route is more suitable for long-tail applications willing to open permissions and is unlikely to become a universal solution for connecting leading super apps.

The second route is the native API collaboration model: AI assistants rely on standardized interfaces to invoke platform services. This solution has stable interaction, controllable permissions, and clear responsibilities. Platforms can independently choose the scope of capabilities to open and independently design profit-sharing rules.

The cost is a slow pace of advancement, requiring business negotiations one by one, finalizing interface specifications, and negotiating revenue sharing. Moreover, most leading platforms already have their own AI business layouts and may not be willing to open core interfaces to external AI assistants.

The industry is more likely to move toward an intermediate form: some applications with an open attitude open native interfaces, and AI undertakes high-value tasks in safe and stable scenarios; for leading super apps that refuse openness, AI can only identify task blockers, prompt users to intervene manually, and after completing the jump, no longer automatically execute subsequent actions.

The narrative of an all-powerful Agent that handles all matters with one sentence may have its applicable boundaries continuously compressed by ecosystem reality. Phone manufacturers previously often depicted product visions with cross-application full-link operations. After the SAEP incident, this narrative needs recalibration: the obstacle does not come from technical capability, but from constraints at the ecosystem level.

For mobile AI agents to become deeply embedded in users' end-to-end operations, three core stakeholders are indispensable: users seeking efficiency, AI developers aiming to push the boundaries of capability and unlock commercial value, and platforms determined to preserve their dominant control over traffic and transactions. The interests of these three parties are inherently in tension, and existing protocol frameworks establish only basic rules without addressing the most fundamental issue: how interests are to be distributed.

When a platform opens its permissions, it is in essence transferring part of its control over user decisions. Without an equivalent mechanism for profit-sharing, traffic exchange, or value distribution, such a transfer of permission lacks commercial justification.

Over a longer horizon, competition in the second half of the mobile AI agent race may no longer hinge on technical metrics such as model parameters and computing power, but rather on whether a collaborative solution acceptable to all parties can be found within the existing closed-loop traffic ecosystem.

Technology can produce agents that read screens and execute instructions, but enabling them to reliably handle affairs for users within the real mobile ecosystem is fundamentally a commercial proposition: what do platforms stand to gain by opening this door of permissions?

The computing power race will continue, but what ultimately determines the ceiling of mobile AI agents' development is no longer computing power itself.

What the mobile Internet has built was never a technical wall but a commercial one. As long as the problem of income distribution behind that wall remains unsolved, even the most formidable computing power will struggle to penetrate the traffic closed loops of super apps.