A snapshot of the fierce competition.

Let's begin with this scene: on the eve of the National Day holiday, a piece of news about ByteDance's Doubao employees "working overtime" went viral. A blogger on a social media platform revealed that an internal team would not take the Mid-Autumn Festival off, launching a high-priority emergency new project to rapidly rebuild a product targeting Meta Muse. The pace was reportedly swift: joint debugging on September 27, with a trial version required by September 30 for internal testing, a total project cycle of only about 6 days, amounting to lightning-fast development.

This meant multiple upstream and downstream teams had to cancel their holidays and go full throttle. Once again, we felt the intense AI race among big tech companies. As the National Day holiday arrived, many projects chose to concentrate their firepower during the long break. This is the most vivid scene of the AI era.

Overtime, overtime. For big tech, this year's long holiday was far from peaceful. Behind this, Muse's explosive popularity was obvious to all. After its launch, it quickly topped the free app charts on North American app stores, surpassing 2.5 million downloads within just over ten days, growing faster than ChatGPT's initial version, and directly driving a significant rise in Meta's market value. What truly set it apart was not a leap in model capability, but turning the personal AI Agent from a laboratory concept into a consumer-grade product that ordinary users could directly pick up and use.

After that, the market's feedback signals were extremely rapid. Information circulating on social media showed that to catch up with the competing project, Doubao's entire product line was mobilized, along with multiple upstream and downstream support teams covering core modules such as algorithms, engineering, product, and design, with some employees working overtime starting from the Mid-Autumn Festival. A viral post on social platforms said: "For the first time working overtime at ByteDance for triple pay, it's incredibly satisfying."

More details emerged: reports said that in April this year, Doubao internally launched a beta project codenamed Spell. After Muse's explosive rise, ByteDance upgraded the Spell project's capabilities and accelerated development, planning to strip these capabilities from hardware limitations and turn them into a standalone consumer-grade personal assistant app usable on all device models, reportedly named "Xiaodou" internally, directly targeting Muse.

As AI enters a high-speed iteration cycle, concentrated sprints during holidays have become a realistic state of the industry. This brings to mind the AI battle during this year's Spring Festival, when big tech employees collectively worked overtime in a similar scene. At that time, approaching the Spring Festival, a Tencent Yuanbao staff member involved in overtime told the media: they worked overtime during the Spring Festival to support Yuanbao's Spring Festival activities, and on the other hand, to wait for the DeepSeek V4 model update. Tracing back to late January, Pony Ma announced at the company's annual meeting that its AI application Tencent Yuanbao would launch a Spring Festival campaign distributing 1 billion yuan in cash. No one dared to let their guard down.

ByteDance was also part of the large group working overtime during the Spring Festival — out-of-town teams rushed to Beijing to stay behind, supporting Doubao during the holiday. AI swept through, and the fiercest Spring Festival battle in history began.

There was also Alibaba (BABA). As Alibaba Chairman Joe Tsai previously revealed, after DeepSeek released the R1 model that year, Alibaba realized it had fallen behind in the AI field. So Alibaba quickly decided: "Cancel the Spring Festival holiday, everyone stays at the company, works overtime and sleeps in the office. We need to accelerate development." Within weeks, Alibaba launched the Qwen series of models.

After a vigorous collective "war," we saw — once the 2026 Spring Festival passed, the AI era truly arrived.

Big Tech AI Departments Go Full Throttle

The AI competition is becoming increasingly fierce. Looking around, AI has entered an ultra-fast iteration period. After more than two years of widespread adoption, user growth for general conversational large models has gradually entered a plateau, and pure Q&A interaction alone can hardly build a sufficiently strong product moat. Industry consensus has shifted — the market no longer needs just AI that excels at chatting. The next generation of AI competition is no longer about conversational experience, but about digital agents that can autonomously handle complex tasks. Simply put: no more casual chatting, but doing things for you.

Thus, a new round of arms race has begun. Recently, big tech companies' moves in personal AI Agents have become noticeably more frequent. Especially after Meta's Muse went viral, it further pushed tech giants to accelerate and deliver their own answers, with emergency project launches and another round of encirclement arriving swiftly.

For example, Tencent quietly launched its cloud-resident personal Agent LightVela. This is a cloud Agent hosting platform built by Tencent Cloud's lightweight cloud team, featuring 7×24 uninterrupted operation, long-term memory, scheduled tasks, and installable skills. It is said that in venture capital circles, Tencent has already earned something of a "Chinese version of Muse" reputation with LightVela.

A few days ago, Manus released its 2.0 version for overseas users and launched Cue, an intelligent assistant for personal life scenarios, equipping the Agent with an independent email and virtual identity, capable of handling personal affairs within a user-authorized budget. Manus, which has been operating overseas for over a year, is currently preparing its domestic version and team, and cooperation with domestic model manufacturers is also underway.

Alibaba (BABA) formally announced its Qwen Personal Agent plan at the Yunqi Conference, relying on consumer ecosystems such as Taobao, Alipay, and Amap, hoping to create an exclusive digital assistant for ordinary people, connecting personal consumption and lifestyle data under user authorization.

Overseas tech giants are also pressing ahead with product iterations. For instance, Musk's xAI launched Grok Bot on August 12, nearly a month earlier than Muse, but only for paying users without breaking into the mainstream. However, none of the companies have fully closed the commercialization loop. Just as Muse achieved phenomenal breakout success, it encountered API blockades from third-party platforms, high computing costs from dedicated virtual machines, and still-high failure rates for complex tasks... Each is a hurdle on the path to personal Agent implementation.

Under the fierce battle, manufacturers at home and abroad still face common real-world challenges.

Every Long Holiday Is a Battle for Users

For big tech companies, holidays are also battlefields. In the mobile internet era, holidays were mostly about operational campaigns to drive traffic; in the AI era, they are a must-fight period for backend R&D teams to do version iterations, tackle technical challenges, and catch up on track gaps. When the outside world enters vacation mode, teams can detach from the complicated daily schedule and concentrate firepower on key technical bottlenecks and advance key projects.

Thus we see that from Spring Festival to Mid-Autumn to today's National Day, big tech companies that like to strike first are sprinting at full speed during holidays. Giving up vacations combined with high incentives reflects the industry's anxiety over fleeting track windows. Today, the iteration cycle of AI products is greatly compressed. Once a new product form validated by the market emerges overseas, the reaction and catch-up time left for domestic players is very limited. If the precious positioning opportunity is missed, by the time user habits and industry patterns are initially set, catching up often requires several times the manpower, time, and resource costs.

But everyone is also clear-headed: high salary incentives and holiday sprints are means for big tech to seize the initiative, but not the winning card. The game in the personal AI Agent track is never simply about development iteration speed. Even if overseas products' surface functions can be quickly replicated, the many real-world tests cannot be bypassed. Zooming out to competition across the entire AI track, what matters is not just development speed, but also ecosystem integration, risk control capabilities, and long-term product determination.

As holidays pass amid code and debugging, in this AI race with no pause button, the real tough battle is still ahead. Everyone is rushing forward with all their might.