Wang Yizhou believes three types of applications are most likely to "survive" in the future. "80% of apps will disappear," stated Peter Steinberg, founder of the open-source AI agent OpenClaw, in a recent interview. In his view, the essence of most apps is data management, and personal agents can take over these tasks and perform them better. He has already enabled a personal agent to adjust fitness plans based on sleep and stress levels, control home lighting, and regulate the temperature of a smart bed. Personal agents possess more information than any single app and make more rational decisions. Previously, Tesla Motors CEO Elon Musk proposed an even more "radical" timeline: within the next five to six years, operating systems or applications will completely vanish. "Your phone just displays pixels and makes sounds. It predicts what you most want to see and hear, then generates it in real-time. We will integrate AI into this device as much as possible," he said. Several industry insiders indicated that the predictions by Steinberg and Musk are not alarmist but reflect a reality already unfolding.

"The Intention Economy" In fact, people have gradually begun experiencing what "fewer apps" feels like, such as ordering milk tea via an AI command or using OpenClaw through common social media platforms without downloading new apps, issuing commands and receiving feedback seamlessly. Tan Yinliang, a professor specializing in AI economics and management information systems, stated that in the age of agents, AI will be the sole "super gateway." Apps will devolve into service-providing plugins (APIs), responsible only for fulfillment. Musk's concept of the "disappearance of the operating system" resembles more the "disappearance of the operating system's presence"—the underlying system remains, but the primary interface shifts from icons, menus, or windows to "Agent dialogue + automatic execution." This transformation in human-machine and service interaction means that development centered on user interfaces (UI) will face disruption, as agents do not need to "browse" pages or "click" buttons like humans. "Previous development revolved around the user interface, with the frontend connecting to users and the backend to functions. But now, all development focuses on intention and execution," said Wang Yizhou, a partner at a management consulting firm focused on hard technology. He emphasized that the future will prioritize whether AI can accurately analyze user intent, distribute commands, and solve problems, highlighting the importance of collecting personalized user data. Tan Yinliang also noted that this shift is disruptive for professionals. Developers will focus more on backend API response speed, the quality of structured data, and compatibility with large models. Product managers will study dialogue logic and intent recognition accuracy instead of clicks and conversions. The core of product design will shift from "how to retain users" to "how to serve AI agents more efficiently." Correspondingly, metrics like daily active users (DAU) may become less important, replaced by task success rates (TSR).

Winners and Losers in the Agent Era If the vast majority of apps disappear, which are most likely to be among the surviving 20%? What new business services and opportunities will emerge? Steinberg mentioned his desire to provide allowances to agents, paying them for solving problems. Ordering food would utilize delivery services, and there might even emerge services akin to "renting human assistance" for physical tasks. Wang Yizhou suggested that three types of applications are most likely to survive: social software capable of handling complex human intentions, services like ride-hailing and food delivery that require offline fulfillment loops, and creative tool applications that serve as carriers and extensions of human will without preset specific intents. Notably, the importance of smart hardware is repeatedly emphasized. Steinberg believes that only apps relying on hardware sensors like cameras and GPS for real-time data collection might endure. Wang Yizhou cited wearable device manufacturers as an example, noting that smart hardware collects real-time data such as eye movements, expressions, and behaviors. These devices not only access data nodes in the real world unreachable by super apps but also supplement behavioral signals to aid intent understanding. "If cloud-based large models handle thinking, then terminal hardware's role is perception and interaction," Tan Yinliang asserted. Future hardware will not passively await commands but will understand the environment in real-time through cameras and microphones. For instance, when a user stares blankly at a refrigerator, smart glasses could analyze its contents and suggest, "You can make scrambled eggs with tomatoes tonight. Although tomatoes are missing, I've already placed an order; they will arrive in 10 minutes." He also stressed that for security reasons, processing users' biometric data and instant speech must occur locally (on-device AI), with only desensitized and compressed intents sent to the cloud, underscoring the growing importance of edge computing nodes. Meanwhile, as apps evolve into APIs, the traditional internet's traffic and advertising models will be the first affected. Tan Yinliang believes future competition will not revolve around selling ad space but profiting from "bidding for priority in AI recommendations" or "service commissions." If users no longer open apps, they won't see splash ads or recommended products in feeds. However, when a user commands "buy running shoes," the AI's decision to recommend Nike or Adidas will become the new commercial battleground. "The pathway, distribution, and matching of intent are far more critical than mere data," Wang Yizhou further analyzed. Traffic previously sold to advertisers might flow in new directions, such as sharing revenue when AI calls third-party plugins to execute tasks, or even being sold to super apps whose traffic is deconstructed and intercepted in the agent era.

"Single Point of Failure" Risks For users, however, security risks are inevitable challenges in an "app-less era." Steinberg emphasized that OpenClaw has access to your computer; if you insist on having it delete all files in the home directory, it may repeatedly confirm with you. If you keep clicking "yes," it will eventually comply and delete itself as well. Tan Yinliang analyzed that when we entrust bank cards, social accounts, and even legal decision-making authority to all-powerful personal agents like OpenClaw, risks increase exponentially. First, security and privacy face "single point of failure" risks; whereas hackers previously needed to breach multiple fortresses like WeChat, Alipay, and email, they now only need to compromise your agent. If an agent is hijacked, it's not just data leakage—hackers can directly manipulate your assets and social relationships. Additionally, if all interactions are AI-generated, humanity could experience the ultimate form of information cocoons. "For tech giants, this is a life-or-death battle over gateways; for ordinary people, it's a complete restructuring of lifestyles. The future winners will be service providers that can make AI understandable and trustworthy," he said.