Português
EnglishDecember 23: AI DingTalk 1.1 New Product Launch & Ecosystem Conference Held
On December 23, the AI DingTalk 1.1 New Product Launch & Ecosystem Conference was held, featuring more than two dozen AI products, including Agent OS, ONE, and DingTalk Real. This marks DingTalk's complete departure from the app-based model of the mobile internet era, with the overall architecture of AI-native DingTalk now fully in place.
DingTalk CEO Chen Hang (nickname: Wuzhao) said: "The 1.0 version released in August was the first step for AI DingTalk. Today, we can announce that DingTalk has undergone a complete transformation, evolving into an AI operating system—a product that reflects what an AI-era platform should look like."
Opening: From 'Fern' to 'Magnolia'
On August 25, we launched DingTalk 1.0, codenamed 'Fern.' Today, we are introducing the brand-new AI DingTalk 1.1, codenamed 'Magnolia.' The name 'Fern' symbolizes deep roots and upward growth, signaling the start of a new AI ecosystem. 'Magnolia,' as the largest plant family in the angiosperm group, combines traditional genetic traits with high plasticity, representing DingTalk's ambition to become a foundational platform for collaboration between countless agents and humans, creating an entirely new ecosystem.
Hardware Launch: DingTalk Real
DingTalk Real is a next-generation terminal device designed specifically for agents. Unlike smartphones or computers that serve human users, DingTalk Real supports the operation of AI agents. It features three core capabilities: Real Entity (authorized access to internal and external systems), Real Data (data retrieval within authorized limits), and Real Time (real-time data acquisition and processing).
The device enables monitoring of agent operations and includes an emergency "power-off" function, ensuring that agents can be intervened if they go out of control. This makes invisible agents controllable and tangible.
System Launch: Agent OS and 'Monkey King'
DingTalk has launched Enterprise Agent OS, building a foundational operating system architecture that supports the operation of various agents. Based on multimodal models, enterprise-owned models, and computing resources, DingTalk has created the Agent Development Platform DEAP and opened up core atomic capabilities such as the MCP protocol, voice intelligence, and visual intelligence.
A new agent manager—'Monkey King'—has been unveiled. As a versatile assistant for users, 'Monkey King' can call upon multiple agents to complete tasks such as message processing, schedule management, AI table operations, research report writing, PPT generation, code writing, inventory inquiries, and automated maintenance requests.
Every day, when users open the DingTalk ONE homepage, a voice assistant summarizes all important information and presents it in a report, enabling one-stop work management. Meanwhile, the 'Discover' feature integrates global information and provides employees with first-hand, cross-industry insights through seven types of agents: New Opportunities, New Dynamics, New Inventions, New Technologies, New Products, New Perspectives, and New Trends.
Vertical Scenario Agents: AI Printing, AI Recruitment, AI Travel
AI Printing offers a one-stop printing service—from digital design to physical delivery—using text-to-image large models combined with human optimization to ensure output meets commercial-grade printing quality, avoiding issues such as text sticking or layout errors. It supports high-quality printing at six central factories, offers same-day delivery in 36 cities, and guarantees data confidentiality at the best prices on the market.
AI Recruitment Agents can automatically collect resumes across the web and communicate with candidates after authorization, confirming interview times and greatly improving recruitment efficiency. For example, in a speed demonstration, two suitable candidates were screened and scheduled for interviews in just 20 minutes.
For campus recruitment, DingTalk has launched "AI Recruitment—Truly Hiring," emphasizing "See real people, apply for real jobs, earn real money." College students can upload 3D resumes via DingTalk, while companies showcase their real office environments and salary packages. The "sincerity fee" paid by companies will be directly disbursed to students after they have worked for three months. The platform takes no commission, promoting transparent connections between employers and job seekers.
AI Travel Agents can automatically compare and book flights and hotels. After authorization, they instantly search for the best prices and complete bookings, helping companies save costs. Employees do not need to manually submit expense reports; the system automatically handles itinerary planning and expense settlement.
Manufacturing Case: Order Tracking and Quality Inspection
In manufacturing, order formats are diverse and often require manual entry, leading to low efficiency and frequent errors. The Order Agent, developed by DingTalk in collaboration with partners, uses document parsing, self-programming, and human-machine interactive iterative learning to achieve 100% accurate order recognition and conversion. The initial learning cost is only 2.5 yuan, and each subsequent conversion costs just 5 cents, significantly boosting efficiency.
In another case, a sensor manufacturing company introduced a time-series model that collects production line data at the millisecond level, achieving 100% comprehensive inspection. Compared to the previous 1% sampling inspection, the missed detection rate dropped from 12% to 1.2%, and the cost of losses fell from 5% to 3.25%, significantly enhancing customer satisfaction.
Guest Speaker: Xu Xiaoying, Youcheng Holdings
Xu Xiaoying introduced that Youcheng Holdings specializes in manufacturing high-precision automotive molds, with its products used in more than 70% of the top 100 car models. DingTalk A1, part of AI DingTalk 1.0, provided real-time translation during overseas meetings, synchronizing information to a domestic knowledge base and greatly facilitating cross-border collaboration.
After the Order Agent went live, the time required to process a single order was reduced from 1 hour to less than 1 minute, and manpower requirements dropped from 7 people per location to about 2 people per location, freeing up staff to focus on higher-value tasks.
Retail and the DEAP Platform Launch
In the retail industry, AI can integrate audio, video, and text data from stores, clean it into AI-ready formats, train industry-specific models, and generate agents such as sales assistants and store manager assistants to aid business decision-making.
Salespeople's AI assistants can analyze service processes and track customer profiles, sending frontline needs directly to the R&D team. Store managers can use agents to evaluate service quality and guide employee training.
DingTalk Enterprise AI Platform (DEAP) has been launched to support enterprise developers in AI data labeling, model fine-tuning, and the construction and deployment of intelligent engines. The billing system is token-based, with enhanced data encryption, prompt protection, and execution auditing to prevent leaks and misuse of computing power.
DEAP helps enterprises advance step by step: L1—improving efficiency through question answering → L2—accumulating knowledge → L3—building business agents → L4—creating proprietary models. In collaboration with the Qwen large-model team, DEAP has launched a cold-start training program that integrates Tongyi industry data with small-sample enterprise data for rapid generalization modeling.
The AI-driven data-labeling system reduces human involvement from 100% to 20%, increasing labeling efficiency by 300%. Model evaluation emphasizes a balance between general and specialized capabilities, using a "cocktail-style" data mix to avoid over-specialization that leads to a lack of common sense.
Guest Speakers: Yisheng Jiankang & Shuimu Molecule
Wang Qiangyu explained that Yisheng Jiankang focuses on women's health medical models and has already launched "Daukou Doctor Super Assistant," which has been integrated into more than 300 hospitals within ten days. The key to medical models lies in "zero hallucination" and "high evidence-based practice," ensuring that diagnoses are supported by scientific literature. With the help of the DEAP platform, the entire chain—from model training to commercial implementation—has been successfully completed.
Xing Jie noted that Shuimu Molecule, incubated by Tsinghua Institute for Intelligent Industry Research, focuses on large models for the pharmaceutical vertical. Currently, AI is being applied in about 30% of new drug development projects. Over the next five years, AI is expected to increase efficiency by 40%–60%, reduce costs by 30%–70%, and drive a transformation in China's innovative drug model.
Open Platform and AI Hardware
DingTalk's open platform has been comprehensively upgraded to support an AI assistant marketplace, an AI hardware marketplace, a curated marketplace, and an AI table template marketplace. These platforms ensure that agent capabilities are more than 90% commercially deliverable and promote a results-based payment model, driving a fundamental transformation in the enterprise-level market.
DingTalk has partnered with Xiaoice Technology to launch an intelligent digital front desk that provides round-the-clock support and proactive guest reception. Together, they have also introduced a "anti-recording magic box," which can be placed in offices to disable recording devices, protecting the security of private conversations and striking a balance between strong listening and robust protection.
A1 and AI Listening
Since its launch, DingTalk A1 has undergone rapid iterations, completing more than 160 optimizations in four months. The model is updated every two weeks, and over 2,000 deep co-creation sessions with customers have been conducted. A1 now boasts powerful listening, transcription, and analysis capabilities.
A new Magnolia limited edition (Xia Yao Red, 500 sets) and a Spring Festival gift edition (999 yuan, 999 sets) have been introduced, making A1 an ideal corporate New Year gift. A1 is evolving from a personal assistant into a team-level work hub, integrating communication data from everyone to provide organizational insights.
AI Listening has fully integrated A1's multi-device audio data and has been upgraded to a multimodal model, supporting automatic analysis of voice, text, video, and images. Content such as photos of whiteboard notes or casual snapshots can be automatically merged to generate written summaries or travelogues.
AI Listening supports simultaneous interpretation—multilingual communication can be achieved with ordinary headphones. During face-to-face conversations, each person's phone can display the other's language; in offline meetings, wearing headphones allows for automatic multilingual translation; and in online video conferences, participants can receive translations in their native languages on demand.
Currently, AI Listening has been optimized for 260 scenarios, becoming a new way of communicating, recording, and learning in the AI era.
AI Tables and a Guest Speaker from YoukeLa
The Yonghe Electronics team used AI Tables to build a Mini MES system in just seven days with two people, achieving end-to-end transparency in production planning management.
A "Universal Sticker" feature has been launched, allowing users to "fill in on the fly," simplifying the data-entry process. The "Snap and Go" feature demonstrates how an Excel table can be converted into an AI application with a single click, completing system creation in just two minutes.
Hand-drawn sketches can also be quickly transformed into usable systems—for example, a sketch of a leave request can be photographed, and a complete leave system can be generated within two minutes.
Wei Jun explained that YoukeLa uses AI Tables to automatically extract keyword hotspots and analyze comments, reducing work that once took hours or even three days to just 15 minutes. This shift has moved the company from "80% data collection + 20% analysis" to "focusing on high-value decision-making."
He urged more enterprises, especially those expanding overseas, to embrace DingTalk's AI, saying, "You must make good use of DingTalk's AI to navigate economic cycles—the future is promising."
AI Search, AI Customer Service, and AI Education
AI Search 1.1 has been upgraded to connect with advanced models worldwide. After a question is asked, multiple models can be automatically called upon for cross-validation and weighted integration, producing multimodal answers and reducing the risk of hallucinations.
A new AI Safety Mode has been added, which automatically de-sensitizes internal company documents and processes sensitive information in an encrypted environment, preventing data leaks during the use of large models.
AI Customer Service 1.1 has enhanced its multimodal understanding capabilities, supporting the recognition of image and video context and integrating the MCP system, which allows direct calls to internal APIs to resolve issues.
The system supports a cold-start mode, enabling it to learn and evolve through daily interactions even without a knowledge base, gradually replacing human customer service representatives.
Upgraded AI Customer Service exhibits strong comprehension, planning, goal-setting, and learning abilities, tracking service tickets throughout the entire process until closed-loop feedback is received.
An AI homework correction machine has been launched, using a copier-style batch-processing approach. It can grade 100 papers every four minutes, covering all subjects from primary to high school, with an accuracy rate exceeding 99%. The machine can handle complex question types, such as detecting short circuits in connection problems or correcting handwriting errors (e.g., extra or missing strokes in characters like "war").
After grading is completed, the system automatically generates statistical reports, including question accuracy rates, class averages, and medians, and pushes the data to parents and teachers. An error notebook is generated automatically, and school-wide student performance data can be stored for teaching analysis.
Tests at Zhuantang Primary School show that grading that once required 106 hours of manual labor can now be completed with a single click. The AI homework correction machine is being offered free of charge to 1,000 primary and secondary schools on a limited basis.
Globalization and Conclusion
DingTalk continues to advance its global strategy, launching "Two DingToks": a China version and an overseas version of DingTalk to meet different regional compliance requirements. The platform supports Google/Facebook logins and has restructured 12 core functions.
The One Office product has been launched, allowing global offices to be unified into a single workspace by connecting them via network telephone lines, with dedicated line support, realizing the concept of "Two DingToks, One Network."
As a final surprise, the entire content of the conference was automatically summarized by AI Listening without any editing, showcasing AI's powerful capabilities in voice and text intelligence.
Chen Hang concluded: "From Windows to Android, the technological paradigm was once defined by the United States. Today, as AI develops rapidly, DingTalk, under the name 'Magnolia,' aims to work with developers, enterprises, and partners to create a new way of working in the AI era—one that belongs to China."
DomTech is DingTalk's officially designated service provider in Macau, specializing in providing DingTalk services to a wide range of customers. If you would like to learn more about DingTalk platform applications, please feel free to consult our online customer service or contact us by phone +852 95970612 or email cs@dingtalk-macau.com. We have an excellent development and operations team with extensive market service experience and can provide you with professional DingTalk solutions and services!