DingTalk CTO Zhu Hong: AI That Doesn’t Enter the Production Stage Is Essentially Just a Demonstration.

On the afternoon of December 23, the 2025 DingTalk Summit · General Manufacturing Session, hosted by DingTalk, was successfully held. The event focused on the transformation of manufacturing models in the AI era, bringing together numerous DingTalk ecosystem partners and representatives from manufacturing enterprises to explore the deep integration and value realization of AI technologies in core manufacturing processes such as order processing, production scheduling, quality control, and process optimization.

DingTalk CTO Zhu Hong: AI must penetrate production processes

"China is a manufacturing powerhouse, and more than 50% of the world's top 500 manufacturing companies are using DingTalk, covering over 30 major categories of manufacturing," DingTalk CTO Zhu Hong noted in his opening speech. He emphasized that for AI to truly take root in manufacturing, it must enter the core production processes of enterprises.

He argued that AI in manufacturing cannot rely solely on general-purpose large models; instead, it should combine industry-specific models with an agile Agent development framework. To this end, DingTalk has created Agent OS—a long-term operating system for AI within enterprises. It integrates computing power, multi-model access, and a low-code development platform, working with ecosystem partners to achieve scenario-based implementation.

Zhu Hong highlighted three core principles: First, put users first, driven by AI to solve real-world production problems; second, enable reuse and support continuous iteration; third, ensure scenarios are implementable—AI must go into the workshop, into work teams, and into production lines.

Taking Youcheng as an example, an order Agent developed on the DingTalk DEAP platform has reduced unstructured order processing time from 1.4 hours to under half a minute, boosting efficiency by hundreds of times. DEAP also introduces a dual-mode paradigm—"development mode + runtime mode"—to meet long-tail customization needs.

In terms of security and data, DEAP supports private deployment, ensuring security through end-to-end encryption, permission controls, and full-link auditing. It also introduces a "data engineering" mechanism that transforms unstructured data into high-quality, AI-ready data, creating an evolutionary feedback loop where the more it is used, the smarter it becomes.

Yizhi Weisi's Yao Chi: Building Zhi Xiao Q, an AI that "understands machine language"

Yao Chi, CEO of Yizhi Weisi, shared the story of "Zhi Xiao Q," an intelligent agent built on the DingTalk DEAP platform, designed to tackle quality control challenges in industrial settings. "Zhi Xiao Q" integrates industrial time-series and vision large models, adopting a "large model + specialized tool plugin" architecture that can understand the physical meaning behind data such as current, voltage, and vibration.

Engineers can simply issue the command "Perform SPC analysis at 6 p.m. every night," and "Zhi Xiao Q" will automatically collect data, generate control charts, and output conclusions. At a Chinese factory of a global top-tier sensor company, "Zhi Xiao Q" has already autonomously completed about 40% of quality engineers' tasks, covering key areas such as process analysis and anomaly detection. All data is deployed locally, ensuring security.

By combining DingTalk Agent OS with industry know-how, "Zhi Xiao Q" has made tacit knowledge explicit and standardized, while also possessing cross-scenario reusability.

Best-practice examples from benchmark enterprises: AI deeply embedded in business operations