The Hard Part of AI Isn’t the Model — It’s the Product

Everyone is racing to add AI features. Very few people are rethinking what products should actually look like in the era of probabilistic systems. That was one of the most interesting parts of my conversation with Manasvi Sharma from Gracenote on the latest episode of *AI Chronicles*. We spoke about: why AI is fundamentally changing product design, not just software architecture the gap between flashy AI demos and reliable production systems why maintainability and evaluation are becoming harder in the age of AI-generated code how multimodal AI systems rely on vectors, metadata, grounding, and knowledge graphs working together why “similarity” alone is not enough without labels and truth underneath One particularly interesting part of the conversation was around multimodal vector databases and graph knowledge systems becoming foundational pieces of modern AI workflows. Manasvi even referenced products like ApertureData as differentiated technologies in this space. Also loved this line from him: “Technology understanding is equal to possibilities.” And yes… we somehow started the episode talking about coffee grinders before getting into AI infrastructure ☕️ Really thoughtful discussion around product thinking, data systems, AI infrastructure, and where all of this is heading. #ai #podcast #artificialintelligence #llm #vectordatabases #metadata