Building an AI proof of concept today is surprisingly easy. With the right tools and platforms, you can get something working in no time.
But turning that PoC into a production system is where the real work begins.
Suddenly, it’s about privacy, scalability, reliability, cost, and automation. Tools that worked fine in a prototype start to show their limits.
In this talk, I’ll walk through what to consider in that transition based on a ‘how-not-to’ and how to move from a quick demo to a robust, scalable AI solution.
I want to talk about the differences between traditional (VMware) and cloud-native (Kubevirt) virtualization. In doing so, I want to show the WoW, the skills, the use case, and possibly the actual implementation. I especially want to give our VMware people a behind-the-scenes look, as this is often still a blind spot.
This is not a session to convince people of KubeVirt, but primarily to show how cool it is in a Cloud-Native environment.
Internally or at customers, we hit the same wall while developing or vibecoding apps with AI components: assembling model serving, MCP and Agent gateways, retrieval, guardrails — months of platform work while AI itself moves weekly. The ITQ AI Use Case Factory is our out-of-the-box solution that clears that wall once, enabling developers to ship apps with Private AI features in a day, not a quarter.
Maxime opens the hood for an honest tour: what’s inside, what we dropped, and what we’d build differently.
