Scaling AI: From Prototype to Trusted Production

AI has crossed from demos into production to agents that remember, reason, and act over time. That shift moves the hard problem off the model and onto the data layer. Durable memory, real-time context, retrieval, and per-tenant isolation now have to work at scale, in one system. In this meetup, Lusha Chen, General Manager APAC at Dify; Tony Huang, Solution Architect at TiDB; and Sean Hong, VP of International Market at Eagle Cloud will share what they've learned taking AI applications from prototype to global scale. Why you must come Talk 1: From Hype to Productivity — The Real Path to Enterprise AI Adoption Every enterprise wants AI. Far fewer have moved past the pilot. Lusha will look at what actually separates the demos from the deployments: where AI projects stall inside large organisations, what has to change in how teams build and govern them, and the patterns emerging from companies that have made the leap from experimentation to everyday productivity. Talk 2: What If Your Product Goes Viral? Inside the Data Architecture Behind Manus and Kimi Manus went viral almost immediately after its March 2025 launch — but its data layer couldn't keep up. Manus had to introduce a waiting list, which soon grew to more than 250,000 users. Manus then migrated to TiDB, a distributed, MySQL-compatible database built for AI agents. Similarly, Kimi K2.6 sits on the same data foundation. In this talk, Tony will walk through the challenges, the trade-offs, and the lessons — and why Manus and Kimi both chose TiDB to run their AI agents. Talk 3: From Shadow AI to Trusted Agents: Practical AI Governance with Eagle Cloud  As organisations rapidly adopt coding agents, autonomous tools, and generative AI, most teams face a growing blind spot: Shadow AI. Employees are already using unapproved tools and personal accounts, while even approved agents often run as black boxes — creating data leakage, compliance, and control risks. In this session, we share a practical framework for discovering uncontrolled AI usage, enforcing policy, and transforming approved tools into governed Trusted Agents with independent identity, least privilege, and runtime protection.  This talk shows how organisations can move fast without creating future security or regulatory debt Agenda