Skip to content

Narrow curated

Pattern: Narrow, Curated Knowledge beats “Index Everything”

Intent
Improve answer quality and reduce costs by scoping retrieval to a small, curated knowledge set aligned to a specific question set.

✅ What Works

  • Be specific. Frame the challenge.
    For example: “We need an Agent to answer our 10 most-asked customer questions, at the right time, using relevant knowledge.”
  • Small, curated knowledge sets (like a single FAQ object) > 100GB of unreviewed docs.

⚠️ What Breaks

  • Expecting an Agent to be an all-knowing oracle.
  • Ingesting massive, unreviewed & unstructured data sets.
  • Skipping pre/post filtering ⇒ poor quality + high cost.

💡 Implications

  • RAG at scale is often fragile and expensive.
  • More data ≠ better answers.
  • A narrow scope often wins: faster, cheaper, more reliable.
  • Overpromising frustrates customers when “AI knows everything” fails.

📌 Guiding Principle

If you wouldn’t train a human with that messy knowledge base, don’t expect an AI Agent to succeed either.