With the release of OpenAI's O-series models and the widespread attention garnered by Deepseek R1's open-source release, competition in the AI industry is intensifying. As large language models continue to improve their complex reasoning capabilities while costs decrease, the industry widely believes that 2025 will mark the beginning of the Agent era.

AI personal assistants (or agents, hereinafter referred to as personal assistants) are considered a special type of intelligent agent, and the industry generally believes they will become the gateway to the next generation of internet. Many major internet companies, AI startups, and phone manufacturers are developing similar products, hoping to secure their place in this future.

This article will provide an in-depth analysis and prediction of the future forms of personal assistant products and key players.

Given the numerous influencing factors and high uncertainty in the future, these predictions may not be entirely accurate. I will attempt to grasp some fundamental principles and extract valuable conclusions for readers' reference.

Key Points:

• The evolution direction of personal assistant products essentially depends on how to maximize AI capabilities.
• The core issues in the evolution of personal assistant products are: how to enable AI to obtain complete context, and how to allow agents to interact freely on the internet.
• The development of AI personal assistants will go through three stages:
  • Multiple top-tier vertical assistants: Existing super APPs evolving into vertical domain AI assistants, this stage won't last long.
  • One comprehensive assistant + multiple professional vertical assistants: Comprehensive personal assistants will provide better user experience. Securing a position in the comprehensive assistant sector means securing a future ticket. Tencent might gain an early advantage, but monopoly will be difficult to form, and WeChat must open its data and interfaces, which will gradually weaken its core moat.
  • Separation of personal assistants and personal data: Users need the freedom to switch between personal assistant products, requiring user data to be relatively independent from personal assistant products. This might lead to the realization of personal data sovereignty, but not necessarily in a web3 form - other possibilities exist.
• Besides existing internet giants, AI phone manufacturers and AI startups are also important forces. Although they may struggle to compete with giants in the existing internet ecosystem,
• agents will reshape internet connection methods. Future agents might achieve direct connections rather than routing through existing internet platforms.
• This will bring major opportunities to reconstruct the internet ecosystem and is the most promising breakthrough point for startups.
• If you're interested in this topic, please contact us using the information provided at the end of this article.

Maximizing AI Capabilities is the Biggest Driver of Personal Assistant Product Evolution

In my previous article "How is Agent Internet Different," I mentioned that the most fundamental question in internet evolution is how to maximize the potential of new technologies.

This principle equally applies to personal assistants: How to maximize AI capabilities determines the evolution direction of personal assistant products. Any factors hindering this direction, regardless of which company or product, regardless of how deep their existing moat is, will eventually be eliminated by the market.

Because in a free competitive environment, if you don't adopt the optimal solution, other companies will inevitably be willing to try and innovate.

So, are the moats built by existing tech giants worthless?

From a long-term perspective, perhaps five years isn't enough, but in ten, twenty, or even thirty years, truly impregnable moats don't exist, otherwise we wouldn't see new giants constantly emerging. If anything can be called a real moat, it can only be continuous innovation capability.

What is the biggest constraint currently limiting the full potential of AI? It's how to enable AI to obtain complete context, and how to allow agents to interact freely on the internet. I believe these are the core essential issues in the evolution of personal assistant products.

Because AI can only make smarter decisions based on complete context; agents can only effectively assist people in completing tasks when they can freely interact with other agents or services on the internet.

If this issue remains unresolved, AI won't be able to realize its true potential, nor provide the best user experience.

Of course, the improvement of AI intelligence level is equally crucial for personal assistant evolution, but this development trend is relatively certain, and we have reason to expect future models that support multimodality and exceed most humans in intelligence level.

Based on this thinking, I divide the development of personal assistants into three stages:

• Multiple top-tier vertical assistants
• One comprehensive assistant + multiple professional vertical assistants
• Separation of personal assistants and personal data

First Stage: Multiple Top-tier Vertical Assistants

When ChatGPT first became popular, the industry's common approach was to develop standalone Chatbot applications, rarely directly restructuring core products. But recently, the industry has started to reflect: Does Chatbot truly represent the future product form?

Of course, some companies like DingTalk and Quark integrated AI closely with their products from the beginning.

After the Deepseek model gained popularity, many companies began integrating it into their core products, with WeChat being the most representative case. This might be a clear signal: The AI-driven restructuring of core internet applications has begun.

As this trend develops, we might see more national-level applications undergo progressive restructuring based on AI.

At that time, multiple top-tier vertical assistants will emerge in the market, capable of helping users complete complex tasks in specific domains using AI, such as planning travel and booking flights, hotels, and tickets through AI, or managing social relationships and information through AI.

The main participants in this stage include several important players:

The first category is existing internet giants, such as BAT, ByteDance, JD, Pinduoduo, Meituan, DiDi, Ctrip, etc.

Their primary strategy is to consolidate their domain advantages (search/e-commerce/travel/social, etc.), restructure core business processes through AI, integrate existing resources, and expand into other domains leveraging their ecosystem advantages.

Tencent might be the strongest in this stage because WeChat's social network moat is so deep, and WeChat is already a behemoth, encompassing not just social networking but also e-commerce, entertainment, gaming, content, etc. While others can't penetrate Tencent's territory, Tencent can strike in all directions to build WeChat into a comprehensive personal assistant.

The second category is AI phone manufacturers.

As important participants in the personal assistant sector, AI phone manufacturers control phone hardware and operating system resources, striving to position themselves upstream of APPs. Currently, no hardware product has emerged that can surpass phones, with earphones and smart glasses existing more as accessories, difficult to become core interaction centers.

From the AI strategies of Apple, Huawei, and other AI phone manufacturers, we can see they generally build their own phone personal assistants on terminals, striving to make them the preferred entry point for user-phone interaction, distributing requests to various APPs through intent recognition.

If this goal is achieved, AI phone assistants will become the upstream entry point for APPs, gaining traffic control rights and thus occupying a more advantageous ecological position.

For this reason, there is obvious competition between AI phones and internet giants. While APPs need the traffic brought by phones, they are unwilling to be completely controlled by phone manufacturers and therefore won't easily open up core user data. Without user data support, AI phone assistants can only complete preliminary intent understanding and distribution, with relatively limited functional value.

The Computer Use feature might be a potential solution, allowing phones to capture user operation processes and obtain user data. However, this solution has shortcomings in user experience and cost (see my previous article: Four Ways AI Interacts with the Internet).

The third category is AI startups, represented by the AI Six Dragons.

These players face a more difficult situation compared to AI phone manufacturers, lacking existing business and product ecosystem support. They can only provide services based on public information or productivity tools, such as creation, Q&A, search, programming, etc., making it difficult to comprehensively penetrate various aspects of users' daily lives.

Will existing internet giants open service interfaces to these startups? I think the possibility is low, as giants won't actively nurture potential competitors.

Overall, personal assistants in this stage can only complete specific domain tasks, and users still need to frequently switch between multiple vertical assistants, playing the role of a "patch worker." Although there might be some degree of data interaction between assistants, this doesn't fundamentally solve the problem.

I predict this stage won't last too long. Large tech companies, AI phone manufacturers, and AI startups won't be content with the status quo, because comprehensive personal assistants can obtain complete user context and provide better user experience. Whoever can first establish a foothold in the comprehensive personal assistant sector will secure a ticket to the future.

Second Stage: Comprehensive Assistant + Multiple Professional Vertical Assistants

Let me first describe the ideal product form of this stage.

With a comprehensive personal assistant, users no longer need to operate between multiple APPs or frequently switch between vertical assistants. The comprehensive assistant can solve various needs in one stop:

• It understands your living conditions and will proactively plan and purchase when daily necessities are running low.
• It knows your preferences and will proactively recommend destinations, plan itineraries, and book flights, hotels, and tickets when you have travel intentions, requiring your confirmation only at the payment stage.
• It helps manage your social relationships and communications, automatically replies to secondary messages, reminds of important matters, and arranges schedules and appointments.
• It grasps your life rhythm and selects music, videos, movies, and other content you love when you need to relax.

Its capability boundaries might far exceed our imagination. But the prerequisite for achieving all this is that it can obtain comprehensive user information and freely interact with other agents on the internet. Only comprehensive personal assistants can truly achieve this.

As for whether professional vertical assistants are needed, for example in medical, health, legal, and other professional fields, I think they might be necessary. But they won't be able to compete directly with comprehensive assistants, existing instead as professional supplements to comprehensive assistants.

Among all players, who is most likely to develop into a comprehensive personal assistant?

Among large tech companies, WeChat undoubtedly has the most potential, still due to its extremely deep social network moat.

At this point, an interesting question arises: Will WeChat open its core interfaces to other assistants, achieving instant message interconnection with other personal assistants?

If not opened, WeChat continues to dominate in a closed state, having such an important product controlled by one company is not good for society, the industry, or even Tencent itself.

If opened, then WeChat's moat will slowly disappear. Once instant message interfaces are opened, social relationship data will eventually need to be opened as well. (Personal assistants might be the most likely product to replace WeChat)

Therefore, I predict that the future comprehensive personal assistant market is unlikely to be dominated by a single player.

If users can freely switch between personal assistants, a key issue inevitably arises: how to migrate user data, and who ultimately owns the user data?

The answer is obvious: Users should have complete ownership of their own data.

If this happens, personal assistants might embrace a new form - the separation of personal assistants and personal data, namely:

• Users can choose to use any personal assistant product.
• Personal assistant products only gain permission to access and write user data, without owning the data itself.
• User data is stored in a node fully controlled by the user.

In this model, personal assistants will become a highly open and active product ecosystem, where anyone can utilize computing resources to create personalized personal assistant services.

Regarding the technical approach to implementing personal data sovereignty, I believe there are two main solutions:

The first is blockchain, but it must solve the blockchain scalability issue.

The second is the Solid project being promoted by World Wide Web founder Tim Berners-Lee, which is a solution that achieves personal data sovereignty using centralized technology.

Of course, technology alone might struggle to drive this transformation; it also needs supporting legal regulations, such as the European Union's General Data Protection Regulation (GDPR).

Overall, the third stage will show the following characteristics:

• Agents replace all existing software
• Agents achieve free connections between each other
• User data separates from personal assistants
• Phones provide computing power and operating system support
• Anyone can develop personalized assistants

This is the perfect solution in my ideal vision.

What I'm uncertain about is, if the core of personal assistants is the AI agent level, what differentiation exists between different personal assistants?

Development Timeline Predictions for Each Stage

The first stage, vertical assistants, can be confirmed to be happening now, and might be completed in a year or two at the fastest.

The second stage, comprehensive personal assistants, is partially happening already. For example, we are designing an open-source agent communication protocol for free agent interconnection. If you have ideas in this area, let's talk (contact information at the end). We believe the internet needs new connection methods and new architecture.

As for the third stage, I'm not sure - maybe three years, maybe five years, but I believe it will definitely happen.

Contact Us

We are developing an open-source agent communication protocol called AgentNetworkProtocol, aimed at achieving direct connections between agents.

AgentNetworkProtocol aims to become the HTTP protocol of the agent internet era.

Our vision is to define the connection method between agents, building an open, secure, and efficient collaboration network for billions of agents.

The ANP open-source technology community is recruiting developers. If you're interested in agent communication protocols, welcome to join us in defining the connection methods between agents and exploring the future of the Agentic Web.

Contact Information: Discord: https://discord.gg/sFjBKTY7sB

Website: https://agent-network-protocol.com/

GitHub: https://github.com/agent-network-protocol/AgentNetworkProtocol

WeChat: flow10240