Free Lesson

Are we all doing "retrieval" wrong?... a candid discussion

Part of Exploring Modern AI Search

60 min

Mar 3, 2026 12:00 PM

Virtual (Zoom)

In this video

What you'll learn

Agentic and Hybrid Search don’t easily “solve retrieval”

Retrieval use cases often need distinctly different data, algorithmic approaches, and UX per domain and per context.

Search is a dynamic ecosystem

Retrieval is way more than picking great embeddings and keyword search tools. It needs critical additional capabilities.

Relevance is a moving target needing ongoing feedback loops

Retrieval engines should constantly adapt to content, domain knowledge, & behavior. They can’t be both static & optimal.

How the experts actually implement retrieval

Real strategies, from experimentation & domain-specific evals, to the algorithms that move the needle on search quality.

Why this topic matters

Modern retrieval usually focuses on static algorithms (BM25, vector, and hybrid search) and models (embeddings, cross-encoders, etc.), decomposing search into isolated features or tools (for an agent). But is this the wrong approach? We’ll discuss why this often hits a wall and when more systematic approaches with continuous feedback loops (user signals, UX, evals, business KPIs, etc.) are needed.