Is the App era coming to an end as OpenClaw gains huge popularity?

Author: Xiang Qing, Editor: Zhao Yuan

In November 2022, ChatGPT was launched. Two months later, its monthly active users exceeded 100 million, making it one of the fastest-growing consumer applications in Internet history.

Many people at that time thought it was just an upgrade of the search method and content production method. But looking back now, what this technological wave has truly changed is probably the operating logic of the Internet itself.

In the past three years, the AI industry has gone through three distinct stages: the model era, the application era, and the upcoming operating system era.

If ChatGPT represents the entrance to large models, then the recently popular OpenClaw gives people hope that AI can transform from a "tool for answering questions" into an "operating system for doing things on behalf of people."

When AI can safely and reliably call tools, access files, operate software, and even actively execute tasks, the structure of future computer systems may also change accordingly.

An AI operating system similar to Windows in the PC era and iOS/Android in the mobile Internet era is gradually taking shape.

I. An Imperfect Enlightener

Different from traditional AI chat tools, OpenClaw can directly operate computers, call software, and execute tasks, which is the key factor for its popularity.

It must be recognized that the current OpenClaw is far from a mature and user-friendly product, and there are even many obvious shortcomings. Its deployment threshold is high, the operation process is not smooth, and it also faces real risks such as permission security, privacy leakage, and rapid Token consumption. These problems make it difficult for it to quickly become a production tool for most ordinary people.

However, the core value of this "imperfect" product lies in completing a crucial industry enlightenment and cognitive breakthrough - allowing more people to intuitively feel for the first time that AI can not only "talk" to give answers but also "do things" to complete tasks.

With the popularity of OpenClaw, domestic technology companies have started the battle for the AI entrance by accessing "Lobster".

In addition to large model companies like Kimi accessing OpenClaw, the actions of tech giants such as Tencent and ByteDance are the most noticeable.

Tencent, which has been relatively cautious in advancing in the AI field, has taken unusually intensive actions this time, releasing five Lobster products in a row, including the desktop AI intelligent agent WorkBuddy, OpenClaw integrated with Enterprise WeChat, OpenClaw integrated with QQ, OpenClaw deployed on Tencent Cloud's lightweight cloud, and QClaw launched by Tencent PC Manager.

More importantly, some of these products can be associated with QQ and WeChat. For example, after installing QClaw, you can directly chat with Lobster on WeChat and let it help you with your work. In the future, when you are suddenly assigned a task by your leader during your break, you can simply send a message on WeChat, and your computer will help you complete the task, including modifying tables, sending emails, and operating browser processes. You no longer have to be interrupted during your break.

Tencent is also promoting an official intelligent agent within WeChat.

According to The Information, Tencent is developing a new AI Agent for WeChat. This Agent will connect millions of mini-programs providing various services within WeChat, covering many fields from booking taxis to ordering groceries, in order to outperform competitors such as Alibaba and ByteDance in the competition. The report says that this project has been listed as a high-priority confidential plan, and it is planned to start the gray-box test in the middle of this year and be officially launched in the third quarter.

ByteDance, Baidu, etc. are also making similar layouts.

Volcengine has officially launched ArkClaw. According to the official introduction, this is an out-of-the-box cloud-based SaaS version of OpenClaw. Without any complex configuration, you can open the webpage and use the 7×24-hour online AI assistant to easily "raise shrimps".

Baidu has also launched the mobile application "Red Finger Operator", extending the capabilities of OpenClaw to mobile devices, supporting users to automate cross-App tasks through natural language instructions and realizing cross-App interactive operations such as taking a taxi and ordering takeout.

Why are these companies acting so quickly?

The core reason is that AI is undergoing a qualitative change from a productivity tool to a system-level entrance. Different from the early chat-style AI, the new generation of AI intelligent agents can call software, operate devices, and automatically complete complex tasks.

If the entrance in the mobile Internet era is the App, then in the AI era, the entrance is likely to become the AI intelligent agent. And the battle for the operating system in the AI era has already begun globally.

On the one hand, AI companies are strengthening the system capabilities of AI.

OpenAI is continuously expanding ChatGPT's tool calling, task execution, and developer interfaces, enabling AI to directly connect to various software services.

Recently, OpenAI launched GPT - 5.4, which introduced native computer usage functions. It enables artificial intelligence agents to interact with operating systems, websites, and applications through mouse, keyboard, and visual input. Developers can use this model to automatically execute multi-step workflows in various software environments.

At the same time, traditional technology giants are carrying out underlying defenses.

Microsoft is deeply embedding AI into the Windows and Office systems, hoping to make AI a new operating entrance; Apple is strengthening the local AI capabilities in iPhone and macOS, trying to integrate AI into the system's underlying layer.

When AI can call applications, operate devices, and execute complex tasks, a new computing architecture is taking shape: user → AI → application service. The competition around this entrance is essentially a battle for a new operating system entrance.

II. The Next Round of AI Competition Depends on Behavioral Data

The popularity of OpenClaw has made Agent one of the hottest directions in the AI industry in a short time. But for technology companies, this competition is closely related to the current real pressure in the AI industry.

In the past few years, large model training mainly relied on publicly available Internet texts, such as encyclopedias, news, books, or forum content. However, as the scale of models continues to expand, the value of these data is declining.

Existing research has pointed out that the growth rate of AI's demand for data far exceeds the speed that real and diverse data sources can provide. The lack of naturally occurring real data is posing a serious risk to the development of artificial intelligence.

The research institution Epoch AI predicted in a study released in 2024 that technology companies will exhaust the publicly available training data for artificial intelligence language models within a decade (approximately between 2026 and 2032).

In the short term, technology companies like OpenAI and Google are competing to obtain high-quality data sources, sometimes even paying for them, to train their large artificial intelligence language models. For example, they sign agreements to obtain a continuous stream of sentences from Reddit forums and news media.

In the long run, new blogs, news articles, and social media comments will not be sufficient to maintain the current development trajectory of artificial intelligence. This will force enterprises to use sensitive data that is currently considered private (such as emails or text messages) or rely on the less reliable "synthetic data" output by chatbots.

The key to improving model capabilities in the next stage is not just more text, but data closer to real behavior.

When a user asks AI to complete a task, AI will go through a series of specific steps, such as searching for information, opening web pages, calling software, or filling out forms. These operations will form a complete task chain, which is what the industry often calls task trajectory data.

Compared with static text, this type of data is closer to the action logic in the real world and has higher value for training AI models with execution capabilities. From this perspective, technology companies are promoting Agent on a large scale also to preemptively compete for the data source for the next round of competition and train their own models.

As more and more users complete tasks through Agent, these operation processes themselves will also generate a large amount of new training data.

In the process of using Agent, users often need to continuously give instructions, correct errors, and adjust task steps. For the AI system, these interaction processes actually constitute a high-quality reinforcement learning data. Every task execution and every correction records the complete trajectory of how AI gradually completes complex tasks.

Once these data are aggregated to the cloud, they may become an important resource for training the next generation of Agent models.

Compared with traditional Internet texts, this type of data not only contains language information but also includes task decomposition, tool calling, and decision-making paths, which has higher value for improving the reasoning and execution capabilities of models.

III. Is AI Entering the "1995 Moment"?

If we look back 30 years, the Internet in 1995 was in a chaotic period.

At that time, the TCP/IP protocol was already mature, but most enterprises were still exploring what the Internet could do, and ordinary people had to face boring instructions to enter the Internet.

It wasn't until the emergence of Windows 95 that it effectively encapsulated the complexity of underlying technologies through a graphical interface and provided a low-threshold creation environment for developers through standardized API interfaces.

This change not only transformed "connecting to the Internet" from a geek behavior to an everyday activity of ordinary people clicking on icons but also promoted the explosion of the PC software ecosystem and kicked off the golden decade of Internet popularization.

30 years later, the AI industry seems to be standing at a similar "1995 moment".

Large models have demonstrated the ability to handle various complex tasks, such as writing reports, generating videos, writing code, analyzing data, operating computers, calling software, and executing tasks. They can do almost everything.

However, in actual use, ordinary users still need to learn complex prompt words and switch back and forth between different web pages and applications to find the right model or Agent to complete tasks.

In other words, AI has sufficient capabilities, but it lacks an organizational center that can transform various AI capabilities into system efficiency.

From this perspective, if Windows 95 was the operating system entrance in the PC era, then the AI era urgently needs its own "operating system". It will become a unified center connecting users, Agents, and application services, including understanding user intentions, decomposing tasks, scheduling tools, and generating results. Users only need to put forward their requirements, and the rest will be automatically completed by the system.

In the past few decades, from Windows in the PC era to iOS and Android in the mobile Internet era, applications have always been the basic unit of the online world. The process of users using mobile phones or computers has always been to open an application and then perform various operations within the application.

However, under the architecture of the AI operating system, this logic may change.

When AI can understand user needs, call tools, and automatically complete tasks, users no longer need to open multiple applications themselves. Instead, they only need to tell AI what they want to do. AI will automatically call different services in the background and return the final result to the user.

In this mode, the structure of the computer system will become: user → AI → application service.

This means that in the AI era, computers may enter a new interaction mode, that is, intention-driven: users no longer need to learn how to use software but only need to express their intentions; the task of the computer system is to understand the intentions and automatically call various tools to complete tasks.

So, in what form will such an AI operating system appear? Currently, the industry is at the intersection of multiple paths of evolution.

One possibility is a new hardware entrance. OpenAI has invited Jony Ive, the designer of the first-generation iPhone, to participate in the research and development of the first artificial intelligence consumer product, hoping that he can replicate the success he achieved when designing Apple's most iconic products such as the iPod, iPhone, and iPad.

According to foreign media reports, this product is positioned as a "third core device" that can be put in a pocket or placed on a desk with a MacBook Pro and an iPhone. Moreover, this device will be small and portable, able to sense the surrounding environment and living situations, and have no screen at all.

Another possibility is to establish an AI entrance on top of super apps. Platform companies such as Tencent and Alibaba are trying to re - integrate their existing application ecosystems through AI, allowing users to call various service capabilities through a single entrance.

Regardless of the form, if this model truly matures, then AI may become the core infrastructure of the next - generation computing platform after PCs and the mobile Internet. Under this new architecture, today's traffic distribution system centered around Apps may also be rewritten, and real commercial power will shift from "application traffic" to "intention distribution rights".