What is prompt engineering?

In practice

The techniques that matter are unglamorous: state the output format precisely, give two or three real examples, tell the model what to do when it is unsure, and separate instructions from user-supplied content so the two cannot be confused. Measuring matters more than phrasing — without a set of test cases and a score, prompt changes are superstition.

Where it fits in a build

Prompting is the first thing to exhaust before considering fine-tuning. It is cheaper, faster to change and easier to audit. Most projects that concluded they needed a custom model had not yet written a careful prompt with good examples.

Common mistakes

• Treating the prompt as a setting rather than as code, so nobody knows who changed it or whether it helped.
• No evaluation set, which makes every change a matter of opinion.
• Mixing user content into the instruction block, which is how prompt injection gets in.

Related terms

• Prompt injection
• Fine-tuning
• Large language model (LLM)

Working on something that involves this?

ASTACKRA designs and builds AI systems, automation and custom software for businesses that need technology shaped around their own workflow. If this term turned up in a proposal and you want a straight answer about whether it applies to your situation, ask us — no obligation, and we will tell you if the answer is no.