Prompt engineering

Prompt engineering is the practice of writing and testing instructions that state the desired outcome, relevant context, constraints, and answer format.

Prompt engineering is the practice of writing and testing the instructions a model receives. A useful prompt states the outcome, supplies relevant facts or files, defines any hard boundary, and names the required answer format, sections, or fields. Suppose an AI must compare two vendor contracts. A compact prompt can identify the clauses to compare, attach both contracts, require citations to the clause text, and request a table. Add role labels or examples when tests reveal a recurring gap that those additions address.

Builder example

OpenAI and Anthropic both recommend retesting inherited prompt scaffolding on current models and keeping additions that repair a measured failure. Long inherited prompts can trigger unnecessary exploration, repeated checks, and inconsistent tool choice. Prompt design should work with source access, tool permissions, and evaluations. A wording change deserves to stay when representative cases show a gain.

You ask an AI to 'write a summary' and get a generic paragraph that could describe any project.

State the intended reader and decision, attach the project update, and request a usable format. Add an example only when representative tests show that the prompt still misses a product requirement.

Common confusion: More instructions can reduce performance. Current guidance from both labs favors clear outcomes and general direction, with narrow constraints for requirements the model would otherwise miss. Examples should earn their space by correcting a measured error.