Zhu Hong, CTO of DingTalk: AI that doesn't enter the production process is essentially just a demo.
On the afternoon of December 23, the 2025 DingTalk Summit - General Manufacturing Industry Special Session, hosted by DingTalk, was successfully held. This event focused on the paradigm upgrade of the manufacturing industry in the AI era, bringing together numerous DingTalk ecosystem partners and representatives of manufacturing enterprises to jointly discuss the in - depth integration and value realization of AI technology in core manufacturing processes such as order processing, production scheduling, quality control, and process optimization.
Zhu Hong, CTO of DingTalk: AI that doesn't enter the production process is essentially just a demo
"China is a major manufacturing country. Currently, more than 50% of the top 500 manufacturing enterprises in China are using DingTalk, covering more than 30 major categories of the manufacturing industry," Zhu Hong, CTO of DingTalk, pointed out in his opening speech. For AI to truly take root in the manufacturing industry, it must enter the enterprise's production process.
For the manufacturing industry to apply AI, a general - purpose large - scale model alone won't solve the problem. The real complexity doesn't lie in the algorithm, but in whether AI can continuously deliver in long - term operation, continuous collaboration, and constantly changing business scenarios. He believes that AI in the manufacturing industry must be a combination of industry - specific models and the systematic capabilities of agile Agent development.
For this reason, DingTalk chose to build Agent OS. It's not just an AI function, but an operating system that enables AI to run in enterprises in the long term. DingTalk provides underlying AI capabilities, including computing power, access to various large - scale model capabilities, as well as low - code and no - code development platforms, allowing enterprises to quickly generate and iterate business applications. Ecosystem partners bring industry algorithms and practical experience to achieve the implementation of differentiated scenarios. Through the "platform + ecosystem" approach, R & D becomes more agile and intelligent, and business innovation within enterprises can be implemented faster and continuously iterated.
Zhu Hong emphasized that there are three core concepts in DingTalk's construction of this operating system: First, user - centric and AI - driven. It's not about using AI for the sake of AI, but to solve real problems in the production process. Second, empowerment. AI is not a one - time project. It needs to be reusable, iterable, and expandable. DingTalk builds platform capabilities together with partners who have in - depth knowledge through the platform. Third, scenarios must be implementable. AI that doesn't enter the workshop, the work team, or the production process is essentially just a demo.
Taking Youcheng as an example, after using the Order Agent developed based on DingTalk's DEAP platform, Youcheng reduced the processing time of unstructured orders from an average of 1.4 hours to less than half a minute, improving efficiency by hundreds of times. At the same time, based on co - creation practices with customers, DEAP developed a dual - form paradigm: the "development state" is responsible for agile construction, and the "operation state" is responsible for efficient execution, thus meeting the long - tail customization needs in the manufacturing industry.
In terms of the guarantee system, Zhu Hong detailed the core design of the DEAP platform in security and data engineering. The platform supports full - fledged private deployment and builds a trusted data security environment through end - to - end encryption, identity authentication, permission systems, and full - link auditing. DEAP also introduces "data engineering", which continuously transforms messy unstructured data into high - quality data that can be directly used by AI through a "machine + human" collaborative model, thus feeding back into the iterative optimization of enterprise - specific models and forming an evolutionary closed - loop of "becoming smarter and more professional with use".
Collaborating with the ecosystem to create the intelligent agent "Zhi Xiao Q" to tackle production quality control challenges
As a deep co - creation ecosystem partner of DingTalk in the field of manufacturing AI, Yao Chi, the founder and CEO of Yizhiweisi Intelligent Technology Co., Ltd., shared how the two sides collaborated to solve the most core quality control challenges in industrial sites. He pointed out that there have long been problems in manufacturing production lines, such as isolated data islands, complex semantics, and traditional analysis highly relying on engineers' experience and professional software. Therefore, based on DingTalk's DEAP platform and combined with middleware capabilities such as AI tables and AI voice recording, Yizhiweisi jointly created the intelligent agent product "Zhi Xiao Q" focused on production quality control.
The core innovation of "Zhi Xiao Q" lies in that it is not a single language model, but a composite intelligent agent integrating large - scale industrial time - series models and large - scale industrial vision models. This architecture of "large - scale model + professional tool add - on" enables AI to truly understand the physical meaning behind time - series data such as current, voltage, and vibration on the production line, dig out patterns in complex production processes, and achieve a leap from "recognizing language" to "interpreting machine language".
In actual implementation, "Zhi Xiao Q" can undertake the daily tasks of quality engineers. For example, when an engineer issues an instruction like "Perform SPC analysis at 6 p.m. every night", the intelligent agent will automatically retrieve data from the MES/ERP system, call the built - in professional tools to generate control charts, and provide analysis conclusions.
Yao Chi took a globally leading sensor enterprise as an example. On the production line of this enterprise's Chinese factory, "Zhi Xiao Q" can already independently complete about 40% of the tasks of production quality engineers, covering key links such as business insight, process capability analysis, anomaly detection, and prediction. All deployments are completed locally at the customer's site, and the data never leaves the factory, ensuring the absolute security of production data. By combining DingTalk's Agent OS platform with the in - depth industry capabilities of ecosystem partners, it is possible to efficiently make the implicit knowledge in the manufacturing industry explicit and standardized, and quickly generalize it to different scenarios, providing a reusable paradigm for solving the long - tail needs of the industry.
First - hand practical experience from benchmark manufacturing enterprises in steel, electrical, photovoltaic, etc.
At the event site, representatives of benchmark enterprises from industries such as steel, electrical, and photovoltaic industries shared their real - world experiences of integrating AI with business from their own perspectives.
Lu Zhaogang, the Party branch secretary and general manager of the Digital and Intelligent Development Center of Liugang Group, emphasized that AI should serve people. Liugang has empowered front - line employees with AI capabilities through two core initiatives: the "Ten - Thousand AI Employees Plan" and the "Complete Set of Digital and Intelligent Tools". By organizing an AI skills competition, employees were inspired to create thousands of intelligent assistants within a week, covering multiple scenarios such as workshops, offices, and sales. At the same time, Liugang used the "Ask for Data" application to drive the standardization of data governance in reverse, and involved AI in daily business such as quality control in team meetings and inspection warnings, promoting the transformation of employees from physical "executors" to intelligent "empowerers".
Li Peng, the IT director of Tianzheng Electric, shared the company's experience in promoting the implementation of AI with a "small - step, fast - run" strategy. Li Peng mainly shared three major achievements realized through DingTalk's AI: He presented a quality management case to show the ability of AI tables to allow business departments to independently create applications. So far, Tianzheng has created several AI table applications and configured more than a thousand automated workflows, greatly improving the company's digital development capabilities; in meeting collaboration, with the help of AI voice recording, the time for generating post - meeting minutes has been shortened to minutes, significantly accelerating the task - closing speed; in sales empowerment, through the AI sales assistant, the time for complex product selection, solution generation, and technical data query has been shortened to seconds, greatly improving the sales response efficiency and accuracy, and promoting the transformation of sales business expansion from an "experience - driven" to a "data + AI - driven" model.
Shen Dongkun, the deputy general manager of IT for planning and architecture at JinkoSolar, elaborated on the "1310" top - level design for AI transformation from a systematic perspective. That is, centered on paradigm change, combined with the enterprise's business strategy, three main lines of business + AI were sorted out, decomposed into ten major fields, implemented in X scenarios, and supported by an AI platform, organization, mechanism, and culture as a guarantee system. He particularly emphasized that AI transformation is not just about technology application, but also "a re - design of the relationship between human and machine intelligence". JinkoSolar promoted the in - depth implantation of AI culture by establishing an AI management committee led by the CEO, organizing digital and intelligent competitions, and establishing a digital talent training and certification system. In the production process, JinkoSolar uses an AI process analysis intelligent agent for root - cause analysis of quality issues; through the collaboration of multiple technologies such as AI, it conducts real - time monitoring of "real - time detection + voice broadcast + DingTalk push" to ensure the standardization of operations, achieving an increase in yield and a reduction in losses.
At the end of the event, five leading manufacturing enterprises, including Sanhua Intelligent Control, Runjian Co., Ltd., Dahua Technology, Greenworks, and Zhongjiate Electric, respectively signed cooperation agreements with DingTalk. They will deepen cooperation in areas such as co - creation of AI intelligent agents, manufacturing site management, global business collaboration, organizational digitization, and intelligent sales, jointly promoting the digital and intelligent transformation of the manufacturing industry and the implementation of AI applications.