---
id: con-prompt-engineering
schema: concept/1
layer: garden
status: fading
name: Prompt engineering
aka: ["prompt hacks", "magic prompts", "prompt formulas", "role prompting"]
summary: >
  The early practice of finding special wording, templates and tricks to coax better answers from
  AI. Much of it matters less as models improve; its lasting core lives on in context engineering.
first_seen: "2022"
relations:
  builds_on: [con-prompting]
  evolved_from: []
  related: []
  parallels:
    - idea: "keyword tricks in early web search"
      field: "search history"
      note: "Quotation marks and special operators mattered a lot until search engines got good at understanding plain questions."
    - idea: "copywriting formulas"
      field: "advertising"
implementations: []
competencies: [cmp-describe-the-task]
origin:
  kind: ai-proposed
  by: "seed agent 2026-10-16"
  rationale: "Shows learners how the field moves: a once-hyped skill fading as its useful core is absorbed elsewhere."
---

In 2022 and 2023, getting good results from AI often depended on wording. People traded templates, told the model "you are a world-class marketer", offered imaginary tips, or added "think step by step". Job ads for prompt engineers appeared. Some of these tricks did help the models of the time.

Newer models understand plain requests far better, and reasoning models already work step by step. Many of the old tricks now make little difference, and some just add noise. The parts that lasted are the unglamorous ones: say what you want, give the background, show an example, describe the output format. Those moved under a wider idea, context engineering, which focuses on what the AI can see rather than how you phrase one line.

The status here is "fading", which means the label is fading. The care it encouraged still matters.

## Example

A marketing coordinator kept a page of "power prompts" from 2023. Testing them against a plain, well-briefed request on today's tools, she found the plain version matched or beat most of them, and was easier for teammates to reuse.

## How it connects

- It grew out of basic **prompting**.
- It **evolved into context engineering**, which keeps the lasting parts.
- **Reasoning models** made several of its tricks unnecessary.

## Try it

Find an old "magic prompt" you or a colleague saved. Predict whether it still beats a plain, clear request. Run both on the same task and compare the results.