What is prompt engineering?

What it means at work

For a non-developer, prompt engineering is less engineering than good briefing. You state the goal, supply background, show an example, set the format and then adjust based on what comes back. Schulhoff et al. note that developers and end-users interact with generative AI systems through prompting and prompt engineering. Teams that do this well keep a small library of prompts that work for recurring jobs, such as weekly reports or meeting follow-ups, so nobody starts from a blank box.

What the sources say

• The field is large. Schulhoff et al. present a taxonomy of 58 LLM prompting techniques and 33 vocabulary terms.
• Examples can matter a lot. Wei et al. report that prompting a language model with just eight chain of thought exemplars achieves state of the art accuracy on the GSM8K benchmark.
• Decide what good looks like first. The Anthropic guide assumes you have a clear definition of the success criteria for your use case. The Anthropic guide also assumes you have some ways to test against those criteria.
• Prompting is not the fix for everything. Anthropic notes that not every success criterion or failing eval is best solved by prompt engineering.

An example

An operations lead writes a prompt that turns raw support tickets into a categorized weekly summary. The first version lumps everything together. She adds the category list, a sample output and the rule ‘flag anything mentioning a refund.’ After two rounds it works, and she saves it for next Monday.

A common misconception

The term makes it sound like a technical specialty, and job ads sometimes present it that way. For everyday work it is a habit anyone can pick up. You do not need code, only clarity about the task and a willingness to iterate.

Sources

• Schulhoff et al., “The Prompt Report: A Systematic Survey of Prompt Engineering Techniques” (arXiv): “Developers and end-users interact with these systems through the use of prompting and prompt engineering”
• Schulhoff et al., “The Prompt Report: A Systematic Survey of Prompt Engineering Techniques” (arXiv): “a taxonomy of 58 LLM prompting techniques”
• Wei et al., “Chain-of-Thought Prompting Elicits Reasoning in Large Language Models” (arXiv): “prompting a 540B-parameter language model with just eight chain of thought exemplars achieves state of the art accuracy on the GSM8K benchmark”
• Anthropic, Prompt engineering overview: “A clear definition of the success criteria for your use case”
• Anthropic, Prompt engineering overview: “Some ways to empirically test against those criteria”
• Anthropic, Prompt engineering overview: “Not every success criterion or failing eval is best solved by prompt engineering”