A case study in a product's core interface changing while its underlying data asset persisted: Glean began in 2019 as enterprise search built on BERT and vector embeddings, then added a chat interface after GPT-3 made conversational retrieval viable, and now frames itself as building a "knowledge graph of your organization" rather than a search box or a chatbot.
The sequencing matters — the search-era investment in indexing and understanding enterprise content became the substrate the chat layer sits on top of, rather than a separate product built from scratch; the interface changed before the underlying value proposition did. Compare Multi-Stage Rocket / Capsule Metaphor for Durable Value (durable value carried across product generations) and Data Network Effect: Interest Graph vs. Friend Graph (a different flavor of data-asset-as-moat).
Reported by Tamar Yehoshua (Glean, President of Product & Technology).
Glean's underlying stack, adopted starting in 2019, pairs Google's BERT models with vector embeddings — applying embedding-based semantic search to index and connect content across an organization's SaaS tools, rather than relying on pure keyword matching. The resulting index behaves as a 'knowledge graph of your organization': every connected SaaS source is indexed and cross-linked, so an assistant built on top of it can function like ChatGPT scoped to a company's own content. Because the knowledge graph was built first, Glean's move from pure search into full chat/agent interaction was a UI layer added on top of existing infrastructure, not a rebuild.