Chapter 5: Connect your data in your AI-first organization
Connect your data
Core ideas
Contrast three retrieval modes: traditional search (user synthesizes), black-box AI (no sources), grounded AI (reasoning over curated corpora with citations).
AI shifts the bottleneck to comprehension speed at the moment of intent. Connect data to the user’s friction point, not only to a search box.
Traditional search is a track meet (pick one best document); AI retrieval is a soccer match (field a coordinated team within the context window).
Strategies layer from context stuffing through filtering, keyword/semantic/hybrid retrieval, and multi-stage pipelines. Combine processes when needed.
Connecting to people and applications introduces handoffs, trust, and write-path integrity (idempotency, retries, throttling).
Principles from the chapter
Treat your context window as a roster construction challenge or fantasy football team. You need “offense” data to achieve the user’s goal and “defense” data to prevent poor outcomes.
Instead of a single approach to field the best context, you can select and combine the best retrieval processes, with each one fielding the context they find most relevant.
Search strategy optimization often makes a greater difference in performance than model and system prompt adjustments.
The critical failure mode for AI-enabled knowledge bases is friction.
AI systems face the same trust deficit as traditional analytics. Transparency is the essential asset to overcome it.
Read the chapter for…
The Napoleon retrieval analogy, mobile banking example, deep dive on each retrieval strategy, and write-access patterns including the investment-assistant scenario.