The Cost of Getting Retrieval Wrong in AI Agents | .local NYC 2026

Subscribe to the MongoDB for Developers YouTube Channel: https://www.youtube.com/@MongoDBDevelopers?sub_confirmation=1 Sign-up for a free cluster → https://www.mongodb.com/cloud/atlas/register Subscribe to MongoDB YouTube→ https://mdb.link/subscribe Most AI failures don't start with the model — they start with retrieval. A recent industry survey found that inaccuracy is now the most-cited AI risk as adoption scales, and for agents that act iteratively on what they retrieve, one bad result compounds into a chain of increasingly expensive mistakes. This session looks at how combining full-text, vector, and hybrid search — running natively on live, operational data — closes that gap, plus a conversation with CVS on what accurate retrieval actually looks like in production. Chapters 00:00 The context advantage for AI agents 02:14 Meet the panel: MongoDB and CVS Health 03:34 Lumina: CVS Health’s enterprise context layer 05:28 Retrieval evolves: BM25, vectors, and reranking 07:02 Why precise retrieval can lower LLM costs 07:50 Simplifying the AI architecture 09:12 MongoDB’s unified retrieval approach 13:32 Cost per query and total cost of ownership 17:47 Demo: automate an executive report with Lumina 20:54 Lumina use cases and reported productivity impact 22:02 Technical, strategic, and adoption challenges 27:14 Building trust and evaluating retrieval quality 31:28 Voyage AI embeddings and reranking 36:43 What’s next: cost, models, and multimodal retrieval 38:22 Audience Q&A 39:26 Prioritizing through bottleneck analysis Visit Mongodb.com → https://mdb.link/MongoDB Read the MongoDB Blog → https://mdb.link/Blog MongoDB for Developers YouTube Channel → https://www.youtube.com/@MongoDBDevelopers