Prompt Engineering is Dead. Long Live Prompt Engineering.

The Old Way

In 2023, prompt engineering meant crafting the perfect paragraph to feed GPT-4. Everyone had tips: “act as an expert,” “think step by step,” “use few-shot examples.” It worked — until context windows grew and models got instruction-tuned.

What Changed

Modern coding assistants don’t use a single prompt. They use a system prompt (you), a task prompt (your request), retrieved context (relevant files), and skill prompts (workflow instructions). Each layer has a purpose:

System prompt  → "You are a senior developer..."
Skill prompt   → "Follow ponytail mode: be lazy"
Context        → "Here's the file: [code]"
Task           → "Add error handling to this function"

Prompt engineering shifted from writing better questions to designing better systems of prompts.

Skills as Prompt Engineering

OpenCode skills are the new prompt engineering. Instead of typing “think step by step” every time, you write a skill file once:

# my-convention.md
When reviewing code, check:
1. Error handling on all network calls
2. No secrets in logs
3. Types are strict

The skill encodes your engineering culture. The AI follows it automatically.

The Skill That Remains

You still need to be clear, specific, and provide context. No prompt — no matter how engineered — can read your mind. But instead of writing one-shot magic paragraphs, you build reusable skill files, provide relevant context, and let the agent chain tools toward completion.

Prompt engineering didn’t die. It turned into software engineering.