News Reporter |
News Editor | Wen Shuqi
On August 3rd, Alibaba’s enterprise-grade Agent product, “QwenWork“ (QwenWork), officially kicked off its public beta.
After hands-on testing, we found that QwenWork comes packed with over a dozen expert kits, covering everything from product design to investment research analysis. Once these kits are installed, users just type in a simple command and get polished results without breaking a sweat.
Take investment research, for example. I wanted to turn a survey spreadsheet into clean, visual data. After uploading the table to QwenWork’s investment research expert kit, I simply typed, “I want to visualize this questionnaire data,” and within seconds, I had a set of neatly formatted charts. For comparison, when I uploaded the same table to general-purpose chatbots like ChatGPT or Doubao, I had to keep tweaking my prompts over and over just to get consistent visuals — a real time-sink.
This product is actually a mashup of three existing tools — QoderWork, MuleRun, and Wukong — and it’s being spearheaded by Chen Yusen, the newly appointed CEO of DingTalk. It’s also the industry’s first Agent product that works seamlessly across desktop, web, and IM platforms at the same time.
In the Agent space, Alibaba was one of the earliest players to dive into R&D and real-world applications, and it’s already spawned several star Agent products internally. With QwenWork, Alibaba is showing off its ability to integrate Agent offerings quickly and efficiently — impressive, especially since Chen Yusen only took the helm at DingTalk and the Wukong business unit less than two months ago.
Why the rush? Because office scenarios have become the latest battleground for tech giants in the AI race. Around the same time, ByteDance and Tencent have also been consolidating their business lines and doubling down on AI-powered office tools.
According to Hu Yanping, a distinguished professor at Shanghai University of Finance and Economics and a researcher in intelligent technology and the smart economy, workbenches have become a must-have for users, and the “battle of a hundred workbenches” is already in full swing. By integrating different modules, features, and products — and connecting user flows with business scenarios, while deepening skills, plugins, data, and delivery capabilities — these workbenches are turning into the super-entry point and application platform for enterprise AI.
Not Another WorkBuddy — Aiming Straight at the Enterprise Market
Unlike the Agent products already out there, QwenWork’s biggest differentiator is its laser focus on the enterprise market.
We’ve learned from insiders that Alibaba currently believes office Agents have already made deep inroads into personal productivity, but there’s still a huge gap when it comes to lifting the productivity of entire organizations.
That said, the insider also made it clear: QwenWork isn’t trying to be another WorkBuddy. “WorkBuddy can only tap into personal context, but in an office setting, the enterprise context matters just as much — and that’s exactly what Alibaba is banking on.”
In plain terms, a generic office Agent only solves individual needs. But turning the needs of a three-person team — or a much larger one — into a standardized product? There’s barely any benchmark for that in the market right now.
That’s precisely why Alibaba pulled together its various Agent teams back in June, channeling the strongest talent and tech into building QwenWork. The core goal: deliver an Agent that’s secure, controllable, and capable of actually doing real work inside a company’s actual business systems.
From a design standpoint, QwenWork is clearly tilted toward exploring enterprise-side AI demands. As we’ve learned, it’s already integrated with DingTalk’s IM, so employees can use QwenWork to summarize group chats, create documents and spreadsheets, send calendar reminders and to-dos, and even handle emails and messages. Down the road, enterprise customers will also be able to hook QwenWork into their real databases and workflows.
Another thing worth noting: on the same day, Alibaba rolled out its next-gen foundation model, Qwen3.8, and plugged it straight into QwenWork. One standout feature of this model is its ability to tackle a wide range of real, professional tasks. In benchmarks, Qwen3.8 could annotate over a thousand relevant clauses across hundreds of legal documents in just one hour, and break down more than 160 hours of game footage frame by frame in mere tens of minutes.
In Hu Yanping’s view, when a model maker internalizes Agent capabilities into its own models — and when Agents can deeply understand and orchestrate the model’s full potential — that’s a recipe for hitting a whole new level of intelligent task performance.
The AI Office Race Among Tech Giants Is Just Getting Started
Over the past two years, the big names like ByteDance, Alibaba, and Tencent have been duking it out over raw model capability. But as model performance starts to converge, the competitive focus is shifting toward user scale and commercial value.
Earlier this year, general-purpose Chatbot products became the main arena for these giants. Doubao, Qwen, and Yuanbao all threw massive marketing budgets at the Spring Festival to win over users. But as user numbers climbed, every product had to face ballooning costs for model inference, compute, and operations — while C-end users’ willingness to pay stayed stubbornly low.
Enterprise office, by contrast, offers clearer paying entities, more frequent use cases, and a much easier-to-measure ROI.
Since March, office has become the new AI battlefield for the tech giants. Alibaba’s QoderWork and Wukong, Tencent’s WorkBuddy, and ByteDance’s TRAE IDE have all become focal points of the competition.
Gartner Research Director Yin Hao told us that when three giants pile in simultaneously, it’s also a fight over who controls the work entry point and the ecosystem. DingTalk, WeCom, and Feishu all have massive existing user bases, and they’re wired into companies’ organizational structures, knowledge, permission systems, and business processes. Those are exactly the kinds of scenario and connection advantages that generic AI assistants can’t easily replicate in the short term.
Looking at the numbers, Tencent has grabbed an early lead in AI office. A report from Analysys shows that as of June, China’s desktop AI-native office agent market had expanded rapidly, with combined traffic surpassing 60 million visits. Of that, WorkBuddy led with 20.97 million monthly visits, TRAE IDE’s domestic version hit 12.79 million, and QoderWork pulled in 7.88 million.
But that doesn’t mean the final landscape for AI office products is set in stone.
In Yin Hao’s view, the real differentiator in the next phase of AI office won’t be about who’s connected to the biggest model. It’s about who can turn enterprise-authorized data, specific application scenarios, and ecosystem connections into safe, reliable, task-executing capabilities.
In this race, industry consensus holds that both model design and Agent engineering are absolutely critical — they fundamentally determine whether a product truly meets user needs and boosts efficiency for both individuals and enterprises.
And according to sources close to Alibaba, the company believes the real gap will be opened by enterprise design and implementation. Because every business has wildly different needs, that’s what sets enterprise-grade office Agents fundamentally apart from personal ones.
Seen from that angle, the enterprise Agent competition isn’t just a tech showdown — it’s a battle of industry insight, organizational understanding, and delivery capability. For all the tech giants, the real game is only just beginning. Models can iterate in the blink of an eye, but truly understanding businesses, accumulating industry know-how, and building up delivery muscle? That takes time.
