Practice

Prompt Engineering

Prompt engineering is the practice of designing, testing and refining instructions for a generative AI system.

Last updated 2026-08-29 · Author

A prompt is an interface between a person's goal and a model's probabilistic interpretation. Prompt engineering makes that interface more deliberate: choose the right context, write an unambiguous task, specify the result, test the response and refine the part that failed. It applies to a one-off question and to prompts embedded in a product or workflow.

What the practice includes

A structure for the work

Gaiish is one way to organise prompt-engineering practice. Its vocabulary puts the questions in view: what is the intent, what context is missing, what instruction should run, which constraints matter, what result is usable and how will it be validated? It does not replace testing, model knowledge or domain expertise.

  1. IntentWhat are you trying to accomplish?
  2. ContextWhat does the AI need to know?
  3. InstructionWhat should the AI do?
  4. ConstraintsWhat rules, limits, tone, format, requirements or exclusions apply?
  5. ResultWhat should the finished output look like?
  6. ValidationHow should the AI verify the result satisfies the request?
Traditional prompt
Make this report better and shorter.
Gaiish prompt
INTENT:
Give a board member the evidence needed to decide whether to continue the pilot.

CONTEXT:
[paste report]
The reader has five minutes and has not seen the underlying data.

INSTRUCTION:
Extract the decision-relevant findings and rewrite the report around them.

CONSTRAINTS:
Do not remove caveats or change numbers. Use plain language and no more than 700 words.

RESULT:
Executive summary, evidence table, risks and decision question.

VALIDATION:
Check every number against the source and list any finding the report does not support.

Why this works better

The structured prompt makes the engineering choices inspectable. It names the reader and decision, supplies the report, limits the rewrite, defines a usable shape and requires source checks. It can improve instruction adherence and reduce ambiguity, while the output still needs review.

Go deeper

Receive the Gaiish key concepts and definitions guide for reference.