Context engineering
Context engineering prepares information for individual model calls: relevant instructions, messages, sources and current states.
Preparing the working context for a model call
Context engineering means deliberately selecting and preparing the information a language model receives for a call. This can include instructions, relevant parts of a conversation, retrieved documents and results from tools already executed. Alongside wording, it matters which information is selected, updated or summarised. A limited context window requires priorities. Persistently stored information is not automatically available to the model on every call; available tool descriptions do not grant additional execution permissions either.
In a fictional service case, a customer asks how their repaired device will be returned. An older conversation summary says collection, while the latest confirmed case state records shipping agreed later. For the answer, the application should supply the current state with its provenance and mark the earlier preference as superseded. A review would check whether the model accurately conveys the agreed return method. Providing more text alone does not resolve the conflict. Explaining the agreed shipping arrangement does not itself create a shipping order.
Sources
Related terms
- PromptA prompt is the input to a generative model. A clear task helps but does not replace data connections or access permissions.
- Prompt engineeringDeliberately designing and checking tasks, context and output rules for a generative model.
- Context windowThe context window limits how many tokens a language model can consider in one processing step. It is not a permanent knowledge archive.
- Retrieval-augmented generationRAG combines information retrieval with generated answers. Access to sources does not remove the need for quality and permission design.
- Function callingA language model requests a described function with structured arguments; the connected application executes it.
- Model cardA model card documents an AI model’s purpose, foundations, evaluation conditions and known limitations for a particular version.
Articles on this topic
AI & Automation with ASES
AI assistants and automated workflows