Economic Observer reporter Ren Xiaoning

Pan Yuefei has recently felt that there is something about TodayAI (hereinafter referred to as "Today"). Previously, he took over a commercial project worth 100000 yuan, but by the agreed payment date, the other party had not received the payment. Last month, the other party called him, stating that there were issues with the contract at the time and that a new process needed to be followed. Both parties had separately agreed on a payment date. On October 20th, Today read his recording and found that the payment deadline had passed. He proactively reminded him to follow up on this outstanding debt.

Pan Yuefei is the head of three small and medium-sized companies, traveling on a weekly basis all year round, and his time is fragmented. After using Today, he believes that this product can basically meet his needs for personal AI assistants.

Personal agents have been one of the hottest tracks in the AI field since September this year. Muse has been online for 12 days and has achieved 640000 daily active users on mobile devices in the United States, nearly three times the initial launch of the ChatGPT mobile application. Instinct's valuation exceeded billions of dollars within one year of its establishment, and within three weeks of its launch, 40% of users actively granted credit card usage rights.

Domestic Internet giants such as Alibaba, Tencent and Byte have also launched or will soon launch similar products. Alibaba's Thousand Questions GO! It was launched on October 13th, and advanced features such as "My Applications" and "Cloud Pro" are currently in the invitation testing stage. The personal agent with the Tencent code name HandyBot has quietly launched a service account on WeChat, with the profile marked as "YourPersonalAIAgent". The operating entity is Shenzhen Tencent Computer System Co., Ltd., and there are plans to launch an independent app in the future. The ByteDance team is also advancing related projects, with a product called Little Bean.

An insider from Qianwen told reporters, Qianwen GO! Although it has not been online for a long time, personal agents are Alibaba's strategic level business; In the Qianwen APP, Qianwen GO! Display alongside the "quick mode" of daily conversations.

Many people believe that personal agents have the potential to become the third commercialized scenario in the AI field, following AI video and coding. Some optimistic entrepreneurs and investors are looking forward to it becoming the 'WeChat of the AI era'.

However, at present, the user base of personal agent products is not large, and Muse, which has gone viral, has accumulated 10 million users in a month and a half. Most domestic products are still only widely spread within the AI and geek circles.

Try Personal Agent

Pan Yuefei first used Today on October 12th, and he is also a beta user of multiple personal agents such as Lingguang, Grok, Dots, etc. Today was discovered through his proactive search, and he found out that it was a product created by Qi Junyuan, the former PC manager of Doubao, so he tried the product.

Pan Yuefei has opened up a large number of personal permissions on Today, covering more than 20 tools he frequently uses in his daily life, such as calendar, email, DingTalk, Feishu, WeChat Reading, Jinshan Document, Tencent Meeting, etc.

In addition, he also installed a third-party Skill on his own and integrated Plaud, an AI recording product. Today, from this, we can know who he meets and what things he talks about every day.

One evening at 11 o'clock, he returned home from a business trip. Today proactively asked him if he had just returned from Shanghai and reminded him that his wife's birthday was approaching.

Calendar is his most frequently used feature on Today. Today will read his schedule information, help him organize and refine his itinerary, and remind him of flight and high-speed rail schedules. Previously, he had to sit in front of the computer, browsing through travel apps while manually entering and cross checking these tasks. Now handing it over to AI for processing, he roughly estimates that he can save 20 minutes per day.

After downloading Today, he uses it an average of three to four times a day.

Today is from a startup company, developed by Hui Chong (Shanghai) Intelligent Technology Co., Ltd., established in December 2025, with Qi Junyuan as the legal representative and a registered capital of 1 million yuan. Qi Junyuan also owns Infinite (Shanghai) Intelligent Technology Co., Ltd., with a registered capital of 10 million US dollars, established in March 2026.

On October 21st, Wuyuan Capital announced that the Today team has completed over $200 million in financing, with investors including IDG Capital, Tencent, Wuyuan Capital, Ant Group, Sequoia China, Transsion Holdings, Tongge Venture Capital, and Shanghai Jiao Tong University Hanyuan Asset Management.

The reporter called Today multiple times to request an interview, but the other party refused the interview citing the need to contact the head of the marketing department.

Today founder Qi Junyuan previously served as the PC head of ByteDance, founded Teambi tion (later acquired by Alibaba), and also served as the head of Alibaba Cloud Drive and Vice President of Feishu Products. Within two days of its establishment, Infinite Company officially announced the completion of a $10 million angel round of financing, with investment from Step Star and IDG Capital, and Step Star providing the underlying model.

TodayAI started internal testing in July this year and officially launched on September 24th. The registered user base has not yet been announced to the public.

In addition to Today, a number of domestic startups have launched personal agent products on the market, including ManusCue, Magic Space Mooli, LinkAI/CowAgent, WishSeen, and others. There are more overseas personal A-agent products, with the most well-known Instinct valued at over billions of dollars. In addition, there are Lindy, Or bits, and some vertical scenario tools: FerryHealth for clinic appointments, Miso for travel planning, and Stanley for content creators.

BAI Capital's senior partner Wang Tianfan has invested in three companies that have deployed personal agents. These companies started research and development six months ago, and their products were first launched in overseas markets. RabbitoOS3 was released on September 22 this year, with both 7-day and 14 day retention rates reaching 60%, and the user base growing rapidly. Wang Tianfan stated that the user base of this product is on par with domestic players such as Today, and each route has its own focus.

Wang Tianfan has long been optimistic about the personal agent track. In his opinion, such products can continuously manage user context and output results that are more in line with needs. More importantly, when the marginal cost of tokens drops to a sufficiently low level, individual agents will unlock a lot of personalized needs. At that time, this will be a bigger track than AI video and coding, and a WeChat level business.

Today's competitors are not just start-up companies, but also large corporations.

At the Alibaba Cloud Summit in September this year, Zheng Sishou, the product manager of Qianwen Business Unit, stated that Alibaba's goal is to turn Qianwen into "your personal agent".

Zheng Sishou said that in the end, individual agents are not just passively executing tasks and catering to users, but gradually forming an ability to understand users' long-term goals, understand users' inner personalities, identify deviations between users' verbal expressions and real interests, and provide reminders and negotiate with users at key nodes.

More suitable for large factories

Personal agents are currently "extremely popular" in the AI industry, but when asked by reporters about several domestic investors focused on the AI field, most of them have not placed bets on this track.

Zhang Yu, founding partner of Qingzhi Capital, said that he has recently seen several entrepreneurial projects that focus on the upstream and downstream of personal agents, and his intuitive feeling is that it is not easy to do and requires huge capital investment. Another investor's judgment is that the platform level opportunities in the AI application layer are likely to belong to large companies.

A top domestic investment institution AI investor started looking at personal agent projects last year. At that time, a group of entrepreneurial teams on the market focused on desktop personal A-agents, most of which were in the internal testing and grayscale stage, with a user base of 100000.

Looking back, these early projects are not having an easy time now. There are several reasons: low user demand and narrow target customer group; The interaction threshold is high, and the technology has not kept up yet; Insufficient willingness of users to pay. He believes that those who can truly create platform level products must have at least experience in leading products or technologies from major manufacturers.

After using Today, Pan Yuefei judged that the best destination for this product may be to be acquired by a large manufacturer. Compared with multiple similar products, he believes that Today has not made any disruptive breakthroughs in product and technology, nor does it have insurmountable technological barriers.

Although Today has helped him solve some pain points, there are still many needs that cannot be met. For example, he hopes that his personal agent product can integrate with high-frequency apps such as WeChat, Taobao, Meituan, Quark Cloud, and QQ Browser, which currently cannot be achieved by Today.

At present, the most connected products of Today are overseas products such as Todoist and Notion, and the ecological coverage of domestic billion level traffic apps is clearly insufficient.

Pan Yuefei hopes to have a personal agent product that can connect to WeChat, as 70% of his high-value private data is stored on WeChat. But WeChat is cautious about such AI automated operation tools. When the Doubao mobile phone was first launched, WeChat was the first app to refuse its intervention; Afterwards, Taobao and Xianyu will also pop up human-machine verification blocking calls, while Agricultural Bank of China and Construction Bank will directly suspend related services.

When the second generation of Doubao mobile phone was released this year, more than 150 apps refused to allow it to perform operations. Personal agents cannot do without ecosystems, which either come from hardware or rely on mature product matrices. In this regard, large companies have more strength. ”The AI investors from the top domestic investment institutions mentioned above said.

Muse's popularity in the United States is closely related to Meta's ecosystem. According to Apptopia data, over 95% of Muse users are also using Facebook, and 63% are using Instagram, indicating a high degree of user overlap.

Among the products that have already been launched domestically, Qianwen GO! Our main focus is to meet the needs of daily life, work, and study scenarios, including providing lifestyle services such as comparing prices and finding discounts, and booking restaurant venues; Editing and generating Office documents, processing data, and other work services; Provide learning functions such as professional academic paper tools.

Among them, "Cloud Pro" supports accessing relevant information from Taobao, Damai, Gaode, and Quark Cloud Drive after user authorization. The above-mentioned Alibaba insider said that currently, Qianwen GO! The main connection is Alibaba's own ecosystem, and in order to connect more ecosystems in the future, it is necessary to negotiate with other companies one by one.

Qianwen GO! The two user cases that have received a lot of attention are: one user can't remember which flight is delayed, so ask GO! Checked 4 orders from 4 airlines, locked China Eastern Airlines MU2445, replied clearly "I definitely refused vouchers and only accepted cash compensation" when customer service proposed voucher compensation, and finally received 70 yuan in cash, which was remitted to Alipay; Another user doesn't know how to apply for Taobao price protection, Qianwen GO! Find the price protection entrance on your own, search for historical orders one by one, and help users recover the price difference of 52 yuan from the price reduction of non stick frying pans.

Wang Tianfan believes that the competitive path for personal agents in China will be different from overseas, and major companies such as Tencent, Alibaba, Baidu, and Xiaohongshu are likely to launch their own personal agent products. But the products that big companies fight for themselves may not necessarily be what users really want. He believes that Doubao, which has the largest user base, is likely to become the biggest winner in this domestic track if it can supplement its context management and calendar scheduling capabilities.

WeChat Level Business

The top investment institution AI investors mentioned above remain optimistic about this track: 'Personal agents are the hub of all information, equivalent to WeChat in the AI era. It is a big business, and we are definitely optimistic about it.'. ”However, who can truly make this product is still unknown.

The difference between this round of personal agents and previous products lies in the significant improvement of AI capabilities, which possess long-term memory and can help users handle more daily chores; Many products are compatible with mobile devices, making them more convenient than desktop versions in the past.

The number of overseas Muse users is still growing. According to Similarweb's estimation, Muse had approximately 2.18 million daily active users on October 5th, one month after its launch. The founder of Instinct mentioned that payments made through Instinct's agency are expected to exceed $1 billion within a year. 50% of consumer commission demand on Instinct is for travel scenarios, and within three weeks, 40% of users actively submit credit card usage permissions.

At present, most personal agent products are not fully open: Muse is only available in regions such as the United States and Canada; Instinct adopts an invitation based system; Qianwen GO! Still undergoing grayscale testing, users are scheduled to queue up.

According to a report by Morgan Stanley, Muse's monthly service cost for a single user ranges from $3 to $130, with a baseline scenario of $37, equivalent to an annual cost of $445, which is already higher than Meta's global user annual revenue of approximately $70. The report also mentioned that Muse's true commercialization on a large scale will not be achieved until the end of 2027 to 2028 at the earliest.

This is a very expensive track, and currently the monthly service cost per user is mostly in the range of $10 to $100, making it difficult for startups to support large-scale free models. Wang Tianfan speculates that personal agent products are likely to take the high-end route, targeting users with strong willingness to pay and providing stronger context processing capabilities.

The AI investors mentioned above also believe that there is currently no product that can ensure a good experience while reducing costs. Whether it is enterprises or users, the cost of computing power is a significant expense. He judged that only when high-quality token fees are reduced to a level similar to 4G data, and everyone's payment for tokens is no longer painful, will the personal agent market experience an outbreak and the AI version of WeChat be possible.