AI Practice18 September 20268 min read
Retrieval-augmented generation has become the default answer to 'how do we make an LLM know our business'. But retrieval is a plumbing layer, not a product. It works when the knowledge base is curated, the questions are repeatable and the cost of a wrong answer is low or reviewable.
Three workflow shapes consistently earn their retrieval layer: support triage over documented policies, engineering knowledge search across repos and runbooks, and drafting assistance grounded in approved templates. In each, retrieval is paired with human review and clear sourcing — the model proposes, people dispose.
Where it does not fit: high-stakes autonomous decisions, rapidly changing data without re-indexing discipline, and anywhere the corpus itself is unreliable. The model can only be as trustworthy as what you let it read.