Summary of Key Points
Agent phones are not merely "phones equipped with smarter AI assistants"; they represent a fundamental revolution in mobile operating systems. They shift the traditional model, where users manually open apps and perform tasks one by one, to a new paradigm where users specify their goals, and the system automatically breaks down tasks, allocates resources, and delivers results. This transformation requires addressing four core issues: task management, capability invocation, memory storage, and rule governance. Additionally, there are challenges related to choosing technical approaches, profit distribution, and security responsibilities. Model companies, smartphone manufacturers, OS platforms, and app developers are all vying for a share of this market, with entrepreneurial opportunities lurking in niche areas such as specialized capabilities and infrastructure.
I. Agent Phones: Not an "Assistant Upgrade," but a Change in Operating System Logic
Traditional phones are centered around apps. For example, to book a business trip, you would need to manually check the calendar for availability, search for flights using an airline app, compare hotel prices, verify travel policies through a reimbursement system, and notify colleagues via WeChat—each step requiring you to switch between apps.
Agent phones, on the other hand, are task-centric. You simply say something like, "I need to attend the WAIC conference in Shanghai next week by 10 AM, within a budget of 2000 yuan, and please let me know before making the payment." The system will then handle everything automatically:
- Clarify ambiguous information (such as the departure location or whether a hotel is needed);
- Access your calendar, company policies, and preferred airports;
- Search for flights and hotels in parallel and plan the itinerary (e.g., selecting a flight first before booking a hotel);
- Retrieve prices using travel app interfaces and only seek your confirmation before proceeding with payment;
- Finally, organize all orders, invoices, and calendar reminders for you.
The key difference: Traditional operating systems manage apps, windows, and permissions (with users typically taking the initiative); Agent operating systems manage tasks, capabilities, memory, and strategies (with the system acting on behalf of the user). This isn't just about adding new features; it's about transforming the phone from a collection of tools into an agent that can assist you with your tasks.
II. The Four Essential Components of Agent Phones: None Can Be Missing
For the system to function effectively, four core components are necessary:
1. Tasks: These are like "to-do items" with various states (created, in progress, waiting for confirmation, failed, etc.) and must be synchronized across devices (e.g., a task initiated on a watch confirmed on the phone or viewed on a PC). Without them, tasks would lose context.
2. Capabilities: Apps are no longer just interfaces that users click on; they should offer callable functions (e.g., checking flights, scheduling appointments, initiating payments). Each function must have clear input and output requirements, permissions, and pricing information—just like a restaurant menu indicating available dishes.
3. Memory: Memory is divided into three categories:
- Work memory (temporary information for current tasks),
- Contextual memory (e.g., previous travel preferences or flight changes),
- Preference memory (e.g., preferred departure locations).
Information must be stored in a way that is traceable and understandable to users, allowing them to see why the system made certain decisions and enabling them to delete unnecessary data.
4. Rules: These determine what the system can and cannot do. For example, flight searches can be automated, but payments require confirmation; schedule changes need lighter verification; sharing contacts should involve authentication. Rules help prevent irreversible actions before user confirmation.
These four components together form the "command center" of the Agent phone.
III. Technical Approaches: GUI, API, or a Hybrid? Which is More Reliable?
There are three main approaches to developing Agent phones:
1. GUI Agents (Simulating Human Interaction): Users interact with the interface just like on a traditional phone—viewing screens, clicking buttons, and entering data. The advantage is compatibility with existing apps, but disadvantages include instability (interface updates, pop-up ads, and verification codes can cause issues), as well as potential bypasses of app security measures.
2. API-Based Approaches: Apps provide standardized interfaces for system calls (e.g., directly accessing flight data without using the app). This approach is more stable and auditable, but it requires apps to actively share their functionality and involves negotiations regarding data usage and profit distribution.
3. Hybrid Approaches (The Best Option): Models handle understanding user intentions, splitting tasks, and handling exceptions; APIs handle high-risk or valuable operations (e.g., payments); GUIs serve as a backup when APIs are unavailable. This approach resembles a project team with a model acting as the manager, APIs as specialized staff, and GUIs as temporary assistants.
IV. Commercial Challenges: How to Allocate Revenue for Each Task?
Traditional phones generate revenue through app downloads, ads, and commissions. With Agent phones, multiple services are involved in task execution, changing the revenue distribution:
- Airlines earn from ticket sales;
- Travel apps charge service fees based on API calls or order processing;
- Travel systems may collect fees to ensure tasks comply with company policies;
- Agents may charge subscription fees or a percentage of completed tasks.
Key principles for fair revenue distribution:
- High commissions should not lead to the recommendation of less favorable options (e.g., more expensive flights);
- Revenue should not be based on the number of calls (the value of checking weather vs. booking changes varies);
- User data must not be sold for commercial purposes; it should only be used for task completion.
If revenue distribution is unclear, users will lose trust, and the entire ecosystem could collapse.
V. Who Will Win in the Age of Agent Phones?
Both giants and startups have opportunities:
- Model Companies (e.g., DouBao, Google): They aim to dominate as system agents, leveraging real-world context and frequent interactions across apps.
- Smartphone Manufacturers (OEMs): They want to become the default agents and control device permissions to avoid being mere hardware providers.
- OS Platforms (Apple, Android): They seek control over rule-making, determining which capabilities can be used and how they are confirmed.
- App Developers: They need to protect their brands, data, and transaction processes.
- Startups:
- Niche Capabilities: Develop specialized functions for complex industries (e.g., contract risk assessment, sales process optimization).
- Task-Oriented Products: Streamline industry workflows (e.g., guiding sales from leads to closed deals).
- Infrastructure: Provide security measures, task evaluation tools, and testing environments to ensure Agent phones work reliably.
- New User Interfaces: Design intuitive interfaces for task confirmation and explanation.
The key to success for startups: Having unique data, rules, or APIs that distinguish their products from generic models.
VI. The Future of Agent Phones
Agent phones will not be standalone devices; they will become central nodes in a personal agent network. Watches can initiate tasks, phones verify user identity, PCs handle complex edits, and connected vehicles receive results. Their success lies in their ability to reliably complete real tasks—without guesswork, overstepping boundaries, following clear rules, and being accountable for their actions. If these goals are achieved, phones will evolve from tools into digital assistants that truly enhance users' lives. Otherwise, they will remain just devices with enhanced capabilities.
(End of the analysis.)