Epilogue: There were many possibilities for things to go wrong

The author shares his experience of presenting a virtual AI-powered speaker at a major event in Blumenau, Brazil. Despite months of meticulous preparation, the team faced significant technical challenges, including an internet outage just minutes before the presentation. While the AI speaker performed well initially, it experienced a mid-presentation failure, forcing the author to intervene and manage the situation live. The article highlights the importance of establishing strict boundaries for AI models, such as limiting knowledge scope and improvisation, to ensure reliability. Ultimately, the author concludes that the project was a success, emphasizing that success is not about perfection, but about achieving objectives through resilience and human-in-the-loop strategies. The piece serves as a reflection on the complexities of deploying live AI systems and the necessity of having a human backup plan for when technology inevitably falters.