Yefan Chen
Product Engineer at Dify
Has long been involved in the design, iteration, and implementation of AI applications and workflow products. He has led multiple Dify product version upgrades and the evolution of core capabilities, continuously advancing AI product practices for enterprise scenarios. Previously, he worked on related products at MetaGPT and is frequently invited to share insights with enterprises and community audiences.
Topic
From Dify 1.14 RC to Enterprise AI Automation: Practical Implementation of Workflows and Sandbox OS Agents
As enterprise AI applications evolve from single-turn Q&A to multi-step execution, simple Prompts and Chat are no longer sufficient for real business scenarios. This talk will combine the new capabilities of Dify 1.14 RC to discuss how enterprises can build an AI workflow system that is executable, governable, and reusable. The session will introduce key capabilities in the new version, including Agent Runtime, Skill, and Sandbox. It will also examine common enterprise scenarios and explain how workflows can collaborate with sandbox-based OS Agents to complete file processing, system operations, context assembly, and multi-step task execution. Participants will see a practical path from scenario identification and workflow design to production deployment and governance, and gain an understanding of how to move AI from demos to production tools that generate real business value. Outline: 1. Why enterprise AI adoption is shifting from Chat to Workflow + Agent 2. Key upgrades in Dify 1.14 RC and their significance for enterprise scenarios 3. Role division and collaboration among workflows, Skills, and sandbox-based OS Agents 4. Best practices for enterprise internal AI applications 5. From PoC to production: methods for building governable, auditable, and reusable systems Audience Takeaways: 1. Understand the capability boundaries of Workflow, Agent, and OS Agent, and learn how to choose the right approach for different tasks rather than blindly adopting “all-Agent” solutions. 2. Learn how to connect file processing, command execution, variable assembly, and multi-step orchestration to build truly executable enterprise AI workflows. 3. Gain hands-on experience and practical approaches for enterprise AI deployment, including permission isolation, failure fallback, human-in-the-loop processes, and governance design. Participants will leave with a practical framework that can be applied to R&D, operations, knowledge processing, and internal automation scenarios.
