Add preferred source
One year after leaving Alibaba, former Alibaba Vice President and DingTalk President Ye Jun has officially entered the enterprise AI Agent space. His new company, Deep Space Sequence, was founded about three months ago and has already secured approximately 10 million yuan (approximately $1.5 million) in funding, led by Guangzhou 4399 Information Technology Co., Ltd., with two paying customers on board. The company chose foreign trade and cross-border e-commerce as its first vertical, positioning itself as a foreign-trade AI work platform covering inquiry responses, marketing ad placement, and customs declaration. Ye Jun said Deep Space Sequence’s ecosystem role is complementary to, rather than competitive with, major tech companies’ office Agent products, and it has applied to become a WorkBuddy ecosystem partner. Meanwhile, Qwen Office launched an e-commerce edition ahead of China’s Double 11 shopping festival, building cross-platform workflows around merchant operations, with e-commerce paying users accounting for over 20%. The two paths reflect a diverging landscape in the AI office sector: large platforms penetrating downward from general-purpose tools, while startups build upward from vertical industries.
Key Elements
One year after leaving Alibaba, former Alibaba Vice President and DingTalk President Ye Jun has finally settled on his entrepreneurial direction: enterprise AI Agents. His new company, Deep Space Sequence, was founded about three months ago and has already secured approximately 10 million yuan (approximately $1.5 million) in funding, led by Guangzhou 4399 Information Technology Co., Ltd., with two paying customers on board.
Ye Jun’s first target vertical is foreign trade and cross-border e-commerce. In his view, this sector is riddled with unsolved problems—for example, helping overseas buyers more precisely match with Chinese suppliers, and using AI to improve response quality to overseas customer inquiries, thereby influencing deal conversion.
Deep Space Sequence’s product positioning is not a single Agent, but rather a foreign-trade AI work platform. The platform handles foundational capabilities such as model orchestration, token planning, memory management, tool invocation, and long-horizon task decomposition, layered with multiple industry-specific scenario Agents covering marketing, ad placement, customs declaration, and more.
In an interview with “Future Human Lab,” Ye Jun said the company currently has a dozen-plus people, primarily composed of product R&D, FDE (Field Development Engineers), and new media marketing staff—there is no sales or marketing team yet. By his own standards, the product is still in the polishing stage and far from mature.
From Big-Tech Platform to Startup Niche
Ye Jun’s departure from DingTalk was somewhat unexpected. On March 31, 2025, he received a role reassignment notice, at a time when DingTalk was fully preparing for its 8.0 launch event, planning to unveil a version that would pivot entirely to an Agent foundation—what he called DingTalk’s “AI version.” That press conference, which never took place, became a regret on Ye Jun’s AI journey.
During his five years at DingTalk, Ye Jun consistently believed that “Agents are the window to the next era,” driving the business toward full AI transformation starting in 2022. He candidly admitted that what he once wanted to do could only be done through a big-tech platform; now, as an entrepreneur, what he can do is leverage his past resources and experience to reach the niches that big companies either cannot or disdain to serve—diving into specific industries and going deep into individual companies’ business processes.
Regarding outside assessments that he was “more of a professional manager, skilled at taking something from 10 to 100 rather than from 0 to 1,” Ye Jun does not shy away. He admitted that if this venture fails, the problem is more likely to lie in the “10 to 100” phase—after all, a startup lacks the resources and talent density of a big company.
Ye Jun said he spent nearly 19 years at Alibaba, working on cross-border e-commerce in the early years and enterprise services for the long haul afterward—experiences that made him deeply familiar with both foreign trade and enterprise-grade business. Only after his non-compete period ended did he formally incorporate the company, building the team, finding customers, and seeking investors starting from a team of one.
Product Logic and Business Model
Deep Space Sequence’s business model is top-up plus outcome-based pricing. Beyond fees generated by model usage, the company also charges per specific task—for example, generating a customs declaration form, or producing a video and running ad placements on Meta and TikTok; some scenarios can even be billed based on final exposure or business outcomes. Ye Jun explicitly said he does not want to set fixed subscription fees or per-seat charges—customers pay only when they use the product, and don’t pay when they don’t.
He believes the business model in the AI era is not entirely the same as SaaS. Ultimately, billing should be based on usage and outcomes, and a good business model should be tied to a base that grows organically over time.
In terms of the company’s development roadmap, Ye Jun has laid out a three-step path: Step one is to enter through specific foreign-trade scenarios, using FDEs to understand enterprise business processes, going deep into inquiry handling, customs declaration, and marketing scenarios, then distilling one company’s needs into industry-wide reusable Skills and Agent products. Step two is to build industry-specific models through post-training—putting stable problems that are well-suited for cost reduction via models into vertical models, while leaving fast-changing granular rules at the application layer. Step three is to consolidate capabilities such as model orchestration optimization, memory management, tool invocation, and long-horizon task decomposition into a platform, then expand to other industries.
On the relationship between customization and productization, Ye Jun believes enterprise vertical scenarios absolutely require FDEs and will inevitably involve some customization. The FDE’s job is to translate a customer’s business language into a language that Agents can understand. But one company’s customization needs may not be “custom” within an industry. For example, customs declaration is a scenario every foreign-trade company must handle, and using Agents to help sellers find buyers is a common need. Do it well for one customer first, then distill the common parts into Skills and products that other companies can use directly.
He further explained that AI has lowered the cost of customization—today’s customization is not entirely equivalent to the project-based model of the traditional software era. It is precisely because this requires going into enterprise business sites and understanding industries that large companies are unwilling to do it, while small teams may not be capable of it—and that is where the market niche has formed.
Relationship with Big Tech: Complementary, Not Competitive
Currently, WorkBuddy, Feishu, and Doubao are all pushing hard on office Agents, and WorkBuddy has opened up its ecosystem. Facing these big-tech products, Ye Jun’s assessment is “no competition”—his ecosystem position is to be their complement.
Ye Jun said the office Agent space is well-suited for big companies, and after leaving one, he no longer has the capability to build an entry point. Deep Space Sequence solves industry business processes, not general-purpose office work. WorkBuddy, Feishu, DingTalk, and WeCom can all serve as platforms that Deep Space Sequence connects to—they provide the entry point, foundational capabilities, and relationship chains, while Deep Space Sequence goes into specific industries to understand customer businesses and go deep on processes.
He specifically noted that WorkBuddy made the right call by building an ecosystem. Tencent did not hand the responsibility of AI-era product innovation directly to WeCom, because WeCom and WorkBuddy are two different species. Its ecosystem strategy also fits the To B model—To B moats and compounding returns come from ecosystems, and no single platform can cover every vertical scenario. Deep Space Sequence has already applied to become a WorkBuddy ecosystem partner.
As for Doubao and Feishu, Ye Jun similarly believes cooperation is more appropriate. Internal collaboration within large teams is often slow, and market-based cooperation can sometimes be more efficient—though he stressed this is only his personal judgment.
Qwen Office E-Commerce Edition Launches in Parallel
Just as Ye Jun’s entrepreneurial direction was surfacing, Alibaba’s own AI office push was also accelerating. Ahead of China’s Double 11 shopping festival, Qwen Office officially launched its e-commerce edition, designed for e-commerce operations, with an invite-only beta starting immediately and a goal of full rollout within one to two weeks.
Qwen Office’s e-commerce edition builds a complete workflow around merchants’ daily operations, coordinating cross-platform management with native integrations into Taobao/Tmall’s Business Advisor, Wanxiangtai, Youzan, Qianniu, and other platforms—bringing multi-platform stores into a single table and a unified operations dashboard. The product also closes the loop between “viewing data” and “taking action,” with AI able to execute certain clear-cut rules on behalf of merchants, extending further into brand-level asset production at scale, cross-platform ROI monitoring for ad placements, and competitor activity tracking.
Qwen Office CEO Chen Yusen said at the launch event that he hopes these features will make users realize Qwen Office is not a chat product, but rather a product that can actually get things done when users hand over a complex task.
This is Qwen Office’s first industry-specific edition. Chen Yusen revealed that e-commerce is a business born online—data, transactions, and marketing all happen on the internet—making it almost the ideal industry for Agents to enter. The project kicked off in June this year, with a team of nearly 100 people concentrating development in September to launch ahead of the Double 11 shopping festival.
According to a report by Jiemian News, since Qwen Office’s launch in early August, e-commerce has been one of the deepest-use industries, accounting for over 20% of paying users, with merchants completing an average of 38 complex tasks per week.
Gartner Research Director Fei Tianqi told Jiemian News that industry-scenario Agents differ greatly from general-purpose office Agents, and customers choose products primarily based on ecosystem. Companies whose platform applications and most of their data reside within the Alibaba, DingTalk, and Alibaba Cloud ecosystem will lean toward Qwen Office; companies whose underlying infrastructure uses Doubao’s large models will lean toward Doubao Workspace.
In her view, Agents need to understand context—not just enterprise knowledge, but also what people discussed within the platform and who content was shared with. All of this metadata becomes nourishment for the Agent. Whoever can better plug into a company’s existing data ecosystem will be able to solidify that advantage.
Industry Competitive Landscape and Outlook
The strategic choices of Ye Jun and Qwen Office reflect a notable divergence in the current AI office sector: large platforms starting from general-purpose entry points and platform ecosystems, penetrating downward into industries; startups starting from vertical industries and specific scenarios, building upward toward platform capabilities.
Ye Jun believes that once AI levels the playing field on technical capability, big companies’ talent density and resource advantages are partially diluted. Entering a vertical industry is like entering a jungle—big companies have abundant resources and strong individual capabilities, but they are not necessarily good at fighting a dispersed, long-term battle that requires staying close to customers. Many projects get killed on day one when a strategy meeting sees the market is only a billion-plus yuan, but for a startup, that is already a huge opportunity.
He predicts that To B will ultimately be relatively fragmented, because industry and regional characteristics will give rise to many industry-specific products and platforms. Big companies can easily replicate general-purpose platforms, but they are unwilling to replicate vertical industries. The startup’s opportunity lies in understanding vertical industries and customers, and in providing deep customer service.
For Deep Space Sequence, the first-stage goal is to acquire customers organically online without a sales team, serving 100 unfamiliar paying customers and generating revenue exceeding 10 million yuan (approximately $1.5 million). Ye Jun believes hitting that number means the company has survived.
But the real challenge is scaling: Is the product good enough? Will unfamiliar customers proactively choose it? Is it solving typical business problems? Can it replicate from 10 customers to 100? He admitted he genuinely has no experience with the 0-to-1 phase—”It’s been keeping me very busy lately.”
Qwen Office likewise views Double 11 as a critical window to validate its product value. If merchant feedback is positive, Qwen Office will replicate this product development approach across more industries such as retail, manufacturing, and services.
The ultimate convergence point of the two paths may well be the “niche” Ye Jun described: the deep industry understanding that big companies cannot achieve, and the platform capabilities that small teams cannot build—under the technology-leveling effect of the AI era, these forces are reshuffling the competitive landscape of China’s enterprise services market.
