Qianfeng Interview with Prece Tech: Making AI Truly Participate in Operations and Building a Continuously Evolving AI E-commerce Operating System

【Ebrun Original】Liu Zuohong, founder of Zhejiang Prece Technology Co., Ltd. Prece Tech anchors the e-commerce track, deeply cultivating three major directions: data foundation, AI OS architecture, and vertical business applications. It builds a data semantic layer covering the entire e-commerce link, and with business goals as the core, constructs an integrated system architecture of 'Capabilities + Business Knowledge Base + Skills + Workflows + Business Applications,' driving AI to upgrade from a single-point task tool to multi-agent collaboration that undertakes complete business matters. In this Qianfeng Award interview, Liu Zuohong shared Prece Tech's practices and judgments in the AI e-commerce direction around the industrial modality evolution driven by AI and globalization. Ebrun: This Qianfeng Award takes 'Modality Evolution Brought by AI and Going Global' as its core theme, focusing on the deep reconstruction of industrial capabilities. In the context of the AI e-commerce industry, how do you understand this round of industrial modality evolution jointly driven by AI and globalization? What are Prece Tech's core strategic layouts in AI capability building and global expansion? Could you specifically introduce the company's positioning and business model? Liu Zuohong: AI is currently moving from the first year of Agents into a stage of large-scale explosion. From the e-commerce industry perspective, the changes brought by AI and globalization are superimposed on each other. When enterprises face more markets, platforms, and consumer demands, the operating links and collaboration difficulty will increase; AI gives industry knowledge, operating experience, and cross-link tasks the opportunity to be standardized and automatically reused. For Prece, our current focus is to deepen and thoroughly implement data understanding, professional decision-making, and business execution capabilities in e-commerce operating scenarios, forming a continuously reusable AI operating system. The boundary of AI capabilities has upgraded from single-point tasks and simple Q&A to the replication, automation, and continuous optimization of enterprises' complete business processes; AI OS is also ushering in a key window period for iterative implementation and large-scale commercialization. As industrial scenarios become more complex, the security, stability, and controllability of AI OS become the core foundation threshold, and the industry value logic is reshaped simultaneously—the business model is shifting from pricing based on AI capability invocation volume to value pricing based on business results, continuously accumulating reusable intelligent assets and forming an intelligent compound interest effect. Under this trend, Prece anchors the e-commerce track for strategic layout: deeply cultivating the three major directions of data foundation, AI OS architecture, and vertical business applications, and building a highly accurate and stable data semantic layer covering the entire e-commerce link; with business goals as the core, building an integrated system architecture of 'Capabilities + Business Knowledge Base + Skills + Workflows + Business Applications,' connecting data, intelligence, and business into a closed loop, continuously accumulating reusable intelligent assets in e-commerce scenarios, and realizing intelligent compound interest. Ebrun: You mentioned that what Prece builds is an integrated architecture of 'Capabilities + Business Knowledge Base + Skills + Workflows + Business Applications.' What roles do these five parts respectively play? Why does the e-commerce scenario need such an architecture instead of directly calling general large models? Liu Zuohong: General large models have general understanding and reasoning capabilities, but to deeply root in industrial operating scenarios, they still need dynamically updated vertical industry data and enterprises' real operating data as support. Prece takes business goals as the core and builds an integrated system architecture of 'Capabilities + Business Knowledge Base + Skills + Workflows + Business Applications': Capabilities are the underlying AI capabilities, the business knowledge base accumulates e-commerce operating methodologies, Skills encapsulate callable professional capabilities, Workflows coordinate multiple AI experts around the same business goal, and Business Applications land in specific operating links. Only by connecting data, intelligence, and business into a closed loop in this way can reusable intelligent assets be continuously accumulated in e-commerce scenarios. Ebrun: At present, in which core business links has your company implemented AI? What kind of distinctive AI or digital-intelligence competitiveness system has been formed? Liu Zuohong: Prece's AI capabilities now cover core links across the entire e-commerce chain, including market analysis, product management, content production, advertising placement, and daily operations. Different from single-point AI task tools, our core direction is to drive AI from independently completing scattered tasks to multi-agent collaboration undertaking complete business matters. Around this goal, we have built three core capabilities: first, data understanding capability, enabling AI to uniformly perceive the enterprise's internal operating state and external market dynamics; second, industry expert capability, accumulating mature e-commerce operating methodologies and business experience and encapsulating them into directly callable AI business experts; third, collaborative execution capability, relying on APIs to connect with e-commerce platforms and, under enterprise authorization, completing real business operations, supporting multiple AI experts to work collaboratively around the same business goal. Our differentiated competitiveness is the integration of industry semantic data, e-commerce operating methodologies, platform-controllable execution channels, and enterprise-level AI workflows. Ultimately, we help e-commerce enterprises hand more specialized operating work to AI trusteeship, accumulating a set of AI operating systems that can be reused over the long term and continuously iterated and evolved. Ebrun: You mentioned 'multi-agent collaboration undertaking complete business matters.' How is this different from single-point AI tools? How exactly do multiple AI experts collaborate around the same business goal? Liu Zuohong: Single-point AI tools can only complete scattered tasks, while multi-agent collaboration undertakes complete business matters. We support multiple AI experts to work collaboratively around the same business goal: data understanding capability enables AI to uniformly perceive the enterprise's internal operating state and external market dynamics; industry expert capability encapsulates mature e-commerce operating methodologies into callable AI business experts; collaborative execution capability relies on APIs to connect with e-commerce platforms and, under enterprise authorization, completes real business operations. Only by combining the three can AI move from 'doing tasks' to 'undertaking business matters.' Ebrun: Could you share a most representative innovative practice case of the 'AI value chain'? What common industry pain points did it solve, and what perceptible and quantifiable value did it create? Liu Zuohong: We have already achieved verifiable implementation results in the AI advertising placement scenario. Taking a cooperating large children's clothing brand as an example, relying on Prece AI Business Expert, it achieved fully automated operation of more than 2,000 advertising plans; after implementation in the brand's official flagship store, it achieved million-level incremental revenue in a single week. This case intuitively confirms the value of AI large-scale operations: previously, enterprises were constrained by manpower ceilings and found it difficult to implement mature operating strategies at scale; with AI, standardized professional playbooks can be replicated at low cost and run continuously, supporting business scale expansion. Our core criterion for judging AI value is whether it can bring real business increment and long-term operating efficiency improvement to customers. Ebrun: You mentioned that Prece's Token invocation scale has maintained rapid growth of 16% month-on-month on average since 2026, but it does not simply pursue invocation volume expansion. How does Prece judge whether each Token consumption is worthwhile? What role does the Skill system play in cost control? Liu Zuohong: Prece's Token invocation scale has maintained rapid growth of 16% month-on-month on average since 2026, but we do not simply pursue invocation volume expansion. Through the Skill system, we continuously optimize across multiple dimensions including model selection, link stability, execution response speed, and unit task cost. The core goal is to make every Token consumption anchored to business goals and generate measurable operating value. Ebrun: In the process of advancing large-scale AI implementation, what supporting role do data elements play? Does Prece focus more on deep cultivation of its own industrial data or on advancing cross-entity and cross-industry data collaboration? What are the core challenges, and how do you respond? Liu Zuohong: Data is the core foundation for large-scale implementation of industrial AI. General large models have general understanding and reasoning capabilities, but to deeply root in industrial operating scenarios, they still need dynamically updated vertical industry data and enterprises' real operating data as support. Prece's core data advantages are mainly composed of four core capabilities: first, long-term accumulated e-commerce industry data capability, supporting AI in gaining insight into market trends, competitive landscape, and potential business opportunities; second, data processing capability centered on ontology construction, relying on an indicator system and business knowledge base to jointly build and organize the semantic layer, providing high-quality 'fuel' for AI and completing structured and semantic governance of business data; third, cross-platform operating data access and business execution capability, relying on API channels to aggregate multi-source business data within the boundary of enterprise authorization, and supporting AI in outputting actionable business actions based on semantic understanding, ensuring that AI conducts decision-making and execution based on real operating conditions; fourth, feedback closed-loop capability of the AI execution link, collecting behavioral data and business results generated by AI operations, and reversely evaluating, iterating, and optimizing various vertical AI expert capabilities. In terms of data strategy, we balance the accumulation of industry data assets with the collaborative connection of multi-platform data. At the current stage of implementation, there are still many challenges: data standards vary across e-commerce platforms, interface openness differs, and customers have strict requirements for data security, permission control, and business operation risks. Therefore, data standardization, cross-platform connectivity, fine-grained permission systems, and security risk control are the focus of our continuous investment and construction. We believe that future-oriented data competitiveness lies not in the scale of data stock, but in enabling data to continuously participate in business analysis, business execution, and intelligent capability iteration, building a self-driven data value closed loop. Ebrun: You mentioned data processing capability centered on ontology construction. Why does e-commerce AI need to first establish a semantic layer and indicator system? Why does e-commerce AI need to first build a semantic layer and indicator system? Liu Zuohong: Data processing capability centered on ontology construction relies on an indicator system and business knowledge base to jointly build and organize the semantic layer, providing high-quality 'fuel' for AI and completing structured and semantic governance of business data. In e-commerce scenarios, data such as products, traffic, advertising, users, and transactions are scattered across different platforms and systems, with inconsistent standards and unclear semantics, making it difficult for AI to understand directly. Only by first building a semantic layer and indicator system can data continuously participate in business analysis, business execution, and intelligent capability iteration. Ebrun: How do you view industrial agents rewriting B2B? Do vertical large models have the necessity for long-term development? Where is the AI moat of industrial internet companies? Liu Zuohong: Industrial agents are reshaping the B2B delivery paradigm, driving B2B software from delivering tools to directly undertaking business tasks and delivering business results. Under the traditional model, enterprises purchase systems and rely on personnel operating tools to complete business; in the future, enterprises will more often only need to define business goals, and AI will schedule various professional capabilities for collaborative closed-loop operations and continuously adaptively optimize based on business results. We judge that the evolution of industrial AI will be divided into three stages: the first stage is AI skillification, accumulating industry expert knowledge, experience, and decision logic to form reusable and callable single-item AI professional capabilities; the second stage is AI workflowization, driving multiple AI experts to collaborate and undertake complex, end-to-end complete business tasks; the third stage is the intelligent compound interest stage, relying on business data and result feedback generated by AI execution to continuously feed back into the iteration and evolution of AI experts and workflows, allowing enterprise intelligent assets to continuously accumulate and appreciate as operations run. This is also the main line direction that Prece has long cultivated and continuously implemented. Regarding vertical large models: vertical scenarios cannot do without professional industrial intelligence capabilities, but this does not mean that every industrial enterprise needs to train its own foundational large model from scratch. As the foundational capabilities of general large models continue to strengthen, the core value of industrial players will settle on deep industry cognition, accumulation of mature operating methodologies, compliant business data connectivity, deep execution links on the platform side, and a closed-loop mechanism that can sustainably iterate and optimize intelligent capabilities. Facing the industrial internet, the long-term AI moat is essentially building a self-operating closed loop of industry data, professional decision-making, controllable business execution, and result feedback, realizing the shift from a capability enhancement mechanism to a capability gap amplification mechanism. Ebrun: You divide the evolution of industrial AI into three stages: skillification, workflowization, and intelligent compound interest. From AI skillification to workflowization, and then to intelligent compound interest, what do you think is the most critical condition? From workflowization to intelligent compound interest, what is the most critical condition? Liu Zuohong: The main line direction that Prece has long cultivated and continuously implemented is to move toward the intelligent compound interest stage. From workflowization to intelligent compound interest, the most critical thing is to let business data and result feedback generated by AI execution continuously feed back into the iteration and evolution of AI experts and workflows, allowing enterprise intelligent assets to continuously accumulate and appreciate as operations run. Without real business result feedback, intelligent compound interest cannot occur. Ebrun: What is the biggest challenge in current enterprise AI implementation? How does Prece help enterprises move from single-point applications to large-scale implementation? Liu Zuohong: The main contradiction in current enterprise AI implementation has shifted from the issue of technical feasibility to the difficulty of deeply integrating AI into business processes and organizational mechanisms. Many enterprises do not lack scattered AI tools; what is truly scarce is vertical professional intelligent capabilities that can be directly put into production and use, as well as a long-term operating mechanism that supports AI's normalized participation in daily operations. Merely relying on single-point AI tools to empower individual positions and improve efficiency at single points makes it difficult to replicate and amplify value within the organization. Only by solidifying the mature operating methodologies accumulated by the enterprise into standardized AI capabilities, supporting cross-department collaborative invocation, and continuously iterating and optimizing based on real business, will AI accumulate into an endogenous, long-term usable operating capability for the enterprise. Ebrun: You mentioned that in addition to products, Prece also provides AI workshops and implementation services to help enterprises identify high-value priority scenarios and transform business experience into runnable AI experts and business workflows. How exactly is this service implemented? Why will enterprise AI capability building become a long-term project like digital transformation? Liu Zuohong: Based on the judgment that the main contradiction in enterprise AI implementation has shifted from technical feasibility to business processes and organizational mechanisms, Prece provides AI workshops and implementation services in addition to products: helping enterprises identify high-value priority scenarios, transforming business experience into runnable AI experts and business workflows, promoting normalized use within the organization, and continuously iterating and optimizing the entire intelligent system based on business feedback. We judge that enterprise AI capability building will gradually become a long-term project similar to digital transformation. Different from the relatively static characteristics of traditional digital systems, the AI system can accumulate business experience and self-iterate during continuous use, forming intelligent compound interest. What Prece is committed to delivering to customers is not an AI tool that solves a single task, but an enterprise intelligent system that continuously grows and dynamically evolves along with enterprise operations.       As the annual vane of the TO B field, the '2026 8th Ebrun Industrial Internet Annual Conference and First Industrial Agent Conference' will be held in Beijing on November 20, 2026. The theme is set as 'Modality Evolution Brought by AI and Going Global,' focusing on in-depth deliberation of frontier topics such as industrial agent implementation, AI supply chain value reconstruction, data elements and TOKEN economy, and new ecosystems for industrial going global. The conference will gather academic authorities, benchmark central state-owned enterprises, industrial leaders, investment institutions, and service ecosystem partners, bringing together core industrial forces, deconstructing industrial digital-intelligent transformation paths, and sharing frontier practical experience. At the same time, the annual industrial internet report and benchmark cases will be released, and a decision-maker closed-door meeting and the 'Qianfeng Award' ceremony will be held. This article was first published on the official website of Ebrun.