Let AI write your prompts: the 80/20 approach to prompt engineering

Executive overview

Most people spend time learning prompt engineering best practices — but AI can write better prompts faster than humans can. The smarter move is to use AI tools to generate your system prompts, not craft them manually.

Both OpenAI and Anthropic provide free prompt-generation tools in their developer consoles. Feed them a rough prompt; they output a structured, best-practice system prompt.

Let the model do the prompt engineering — you just supply the intent.

The structure of a good system prompt

• Start with a persona (what role the AI should take)
• Define the task clearly
• Add examples for few-shot learning
• Specify an output format
• Include reminders at the end for things the AI must not forget

How to use the AI-native prompt tools

• Anthropic: console.anthropic.com/dashboard → "Generate prompt" button
• OpenAI: platform.openai.com → Playground → Chat → magic wand icon
• Both require a small API credit top-up (~$2–3, but each generated prompt costs a fraction of a cent)
• Paste your rough prompt → generate → get a structured, improved version

What the tools produce

• Anthropic outputs use XML tags as delimiters — their models are trained to parse XML more effectively than Markdown
• OpenAI outputs use Markdown (hashtags, asterisks) to separate sections
• Both follow the same logical structure: task → steps → output format → examples → notes

Free alternative (no API credits needed)

• Use ChatGPT, Claude, or any free model
• Prompt: "Improve my prompt using prompt engineering best practices. Here is my basic prompt: [paste it]"
• Output quality is comparable — persona, guidelines, format, and few-shot examples all included