Recently, the private sharing session themed “Enabling AI to Truly Understand Enterprises – Enterprise Knowledge System Development and AI Implementation Practices” has come to a successful end. The event gathered more than 60 enterprise managers from the manufacturing and IT sectors to jointly explore how to enable AI to genuinely “understand” enterprises amid the AI wave and revitalize dormant knowledge assets to deliver renewed value.

Distinguished Speakers Gather with Insightful Sharing Addressing Core Industry Pain Points

Seven industry experts from different enterprises were invited to deliver forward-looking and practically-oriented keynote speeches covering intelligent operations & maintenance, centralized data aggregation, digitalization of industry expertise and other dimensions, presenting a feast of professional insights for all attendees.

Ding Zhenxing, Founder of Lerwee, leveraged his 16 years of experience in digital O&M product development and delivered a speech titled Practice of the First Full Toolchain O&M Intelligent Agent. Corporate IT environments have grown increasingly complex, and traditional operations and maintenance are plagued by numerous pain points including difficulties in cross-regional management, fragmented equipment, siloed data, slow fault localization and resource waste. Lerwee pioneered a five-layer architecture consisting of Perception, Planning, Memory, Core Brain and Action to build the Harness Paradigm for O&M intelligent agents, establishing a controllable and credible operation framework supported by four core pillars:

1. Context engineering to dynamically perceive on-site conditions;
2. Architectural constraints to guarantee secure and compliant operations;
3. Feedback loops to drive continuous iteration of intelligent agents;
4. Entropy management to update environmental data in real time and prevent cognitive distortion.

The solution has been deployed in a wide range of real-world O&M scenarios, such as automatic fault diagnosis with sub-second localization and resolution, low-frequency business analysis and cloud resource cost optimization, forming a closed-loop management covering problem identification, analysis and resolution. AI is shifting operations and maintenance from passive response to proactive decision-making.

Mr. Luo from Weike Cloud delivered the opening address before elaborating on eight prevalent pain points in the development and operations of corporate information systems: superficial demand insight, mismatches between business and system logic, heavy legacy architecture burdens, information distortion in multi-layer demand transmission, unclear data management standards, obstacles in troubleshooting system integrations, lack of iteration and accumulation mechanisms, and low efficiency from passive O&M response. Targeting end-to-end pain points, Weike Cloud leverages four intelligent agents to restructure the full process from IT system demand formulation to delivery, unblocking bottlenecks across all links in a one-stop manner.

Yan, Technical Director of Synology, pointed out that in the AI era, corporate data is no longer confined to central server rooms at headquarters, but scattered across production-site terminals, edge hardware, branch offices and diverse cloud platforms. Enterprises are in urgent need of building an integrated full-domain data architecture connecting edge, core and cloud layers. Only unified aggregation of full-domain data can provide sustained data support for continuous iteration and autonomous optimization of AI models.

Mr. Zhou from Camtek shared practical cases. The Pre-CAM phase of PCB manufacturing once relied heavily on the personal experience of engineers. After AI adoption, the system can automatically identify process rules and optimize engineering workflows, generating complete engineering parameter reports within only 5 to 10 minutes. AI is not meant to replace senior industry expertise, but to digitize, standardize and scale scattered experience, converting individual know-how into universal capabilities for the entire workflow.

In response to the data security bottlenecks hindering corporate AI implementation, Inspur presented FusionXpark GB10, a portable intelligent agent development platform that deploys AI computing power on enterprise premises to prevent data leakage. Many enterprises have completed infrastructure construction yet get stuck at the initial stage of AI rollout: AI requires full-domain data to generate value, but sensitive core data including engineering drawings, quotations and customer records cannot be uploaded to the cloud. R&D staff are forced to conduct local AI testing using desensitized data, severely limiting model performance.

Mr. Zhong from Hanma introduced Tuantuan, a post-level intelligent agent platform R&D Workshop for the manufacturing industry. While many enterprises have the infrastructure to run AI, generic tools such as Coze and Dify only deliver superficial functionality and cannot be fully implemented, as they lack industry-specific business rules and data logic for manufacturing scenarios including materials management, approval workflows and customs declaration. Additionally, traditional systems such as ERP and MES suffer severe data silos, forcing employees to consolidate data across multiple platforms, and existing systems lack an intelligent layer to integrate data and generate business outcomes. Unlike platforms built from scratch, the R&D Workshop “Tuantuan” offers a complete implementation path ranging from individual post tasks to overall enterprise capability upgrading.

Finally, Mr. Cheng from Shilietianxia shared insights: many enterprises are fully equipped with storage, computing power and intelligent agents, yet their AI systems still fail to adapt to proprietary business needs. The core bottleneck lies in the fact that files do not equal usable knowledge, and storage does not equate to effective knowledge utilization. The barrier to corporate AI implementation is not model performance, but the inability of AI to interpret internal knowledge. To address this challenge, Shilietianxia has developed a three-tier knowledge architecture:

• Tier 1: Knowledge graphs to unify entity identities, mark timeliness and correlations and eliminate ambiguities;
• Tier 2: Standardized Wiki pages to solidify reasoning conclusions for direct answers to inquiries;
• Tier 3: Vector retrieval for simple queries, with layered scheduling to conserve computing power and balance accuracy and efficiency.

The full-day sharing sessions covered O&M intelligent agents, computing and storage infrastructure, knowledge systems and real-world implementation cases, converging into a unified consensus: corporate AI implementation is far more than simple superposition of individual technologies, but a comprehensive systematic project integrating data, computing power, storage, operations and maintenance, knowledge assets and business scenarios.

Gratitude goes to all speakers and participants for their enthusiastic engagement! Moving forward, Lerwee will continue to partner with industry peers to host more high-quality industry events, building a platform for continuous learning and growth for IT practitioners and supporting high-quality development of China’s IT sector amid the AI era. We look forward to our next gathering to embark on a new chapter of intelligent innovation together!