Context engineering is the apply of designing methods that decide what info an AI mannequin sees earlier than it generates a response to person enter. It goes past formatting prompts or crafting directions, as an alternative shaping the whole surroundings the mannequin operates in: grounding information, schemas, instruments, constraints, insurance policies, and the mechanisms that resolve which items of data make it into the mannequin’s enter at any second. In utilized phrases, good context engineering means establishing a small set of high-signal tokens that enhance the probability of a high-quality consequence.
Consider immediate engineering as a predecessor self-discipline to context engineering. Whereas immediate engineering focuses on wording, sequencing, and surface-level directions, context engineering extends the self-discipline into structure and orchestration. It treats the immediate as only one layer in a bigger system that selects, constructions, and delivers the best info in the best format in order that an LLM can plausibly accomplish its assigned job.
What does ‘context’ imply in AI?
In AI methods, context refers to all the pieces an a giant language mannequin (LLM) has entry to when producing a response — not simply the person’s newest question, however the full envelope of data, guidelines, reminiscence, and instruments that form how the mannequin interprets that question. The full quantity of data the system can course of directly known as the context window. The context consists of a variety of completely different layers that work collectively to information mannequin habits:
