The Context Problem

Because it has no structured knowledge about your business. The model knows the internet. It doesn't know your pricing rules, your compliance requirements, your customer segments, or your operational constraints. Without that context, it generates plausible-sounding answers based on general knowledge, which are wrong in your specific domain.

Partially. Better prompts help, but they're fragile and don't scale. The real fix is structured context: schemas that define your business entities, skills that encode your processes, and context files that capture your domain knowledge. Business-as-Code makes this systematic rather than ad-hoc.

Context engineering is the practice of structuring business knowledge so AI systems can operate on it accurately. It includes defining entity schemas, encoding process skills, and maintaining context that reflects how the business actually works, not how it's documented in a wiki.

More than you think, but less than you fear. A well-structured Business-as-Code implementation for a mid-market company typically includes 20-50 entity schemas, 30-80 operational skills, and domain context covering key processes. This can be built in 2-3 weeks with the right methodology.

RAG (retrieval-augmented generation) helps with factual recall but doesn't solve the structural context problem. Retrieving a document chunk is not the same as understanding your pricing model, your approval workflow, or your compliance constraints. RAG is one tool in the context stack, not the whole solution.