Context Engineering refers to the targeted selection, structuring and preparation of the information that an AI model receives for a query. It is not about the formulation of the instruction itself, but about what knowledge, examples and data are provided to the model before it even answers.
A language model only processes what is in the context window. If the relevant excerpt from the customer history or product documentation is missing, no matter how well-formulated the prompt is, the answer will remain vague or incorrect. Context engineering therefore comprises several building blocks: retrieval-augmented generation for accessing documents, a memory for previous interactions, system instructions that set the framework, and the selection of which information actually fits given limited context windows. This selection is not a technical detail. It determines whether an AI assistant answers from a company's context or from generic training knowledge.
Reference to SAP Business One
Context engineering plays a concrete role in the SAP Business One world as soon as an AI assistant is required to answer questions about documents, master data or reports. A model that has only been trained generally knows neither a client's chart of accounts nor the payment terms of a specific business partner. Only the preparation of master data and document context, for instance via an MCP server or a retrieval pipeline, enables the model to work with the client's actual data instead of guesswork.
Demarcation
Context engineering is not identical to prompt engineering. Prompt engineering optimises the phrasing of the instruction. Context engineering optimises the information available to the model. Retrieval-augmented generation, in turn, is a single technique within context engineering, not a synonym for it. Anyone who only uses RAG, but neglects memory, system instructions and selection criteria, is only doing half of context engineering.
AI in everyday work: Here's how to find the right start
AI assistant meets SAP Business One: Two paths out of the Versino AI shop
Sprint and Swirl: How an AI agent divides work and learns from trials
LATS: Considering several paths simultaneously with AI — and the availability check in SAP B1
ReAct and RLEF: How Artificial Intelligence Learns from Real Mistakes — and What Automated Bank Reconciliation in SAP Business One Has to Do with It
