๐ฆ๐๐ผ๐ฝ ๐จ๐๐ถ๐ป๐ด ๐๐ถ๐ฎ๐ป๐ ๐๐ ๐ฃ๐ฟ๐ผ๐บ๐ฝ๐๐
I started with a simple AI prompt for developer work. It had a role and a task. Then my tasks got harder. Reviewing a pull request is a complex job.
My first reaction was to add more rules. The prompt became too long. One long instruction creates a problem. The AI misses context. The AI jumps to a fix too fast.
I stopped treating the prompt as one text. I started treating it as a process. I split the work into these roles:
- Input intake: Find what is missing.
- Implementation review: Check the solution.
- Action planning: Find the next step.
- Risk review: Check data and permissions.
- Quality check: Find the proof.
- Final editing: Write a short answer.
The user still sees one answer. But the internal work is clear. The AI no longer guesses.
This structure changes the output. Instead of vague tips, you get:
- Blockers: Issues you must fix.
- Questions: Things to clarify.
- Suggestions: Ideas for improvement.
- Conclusion: A clear merge or hold decision.
You get three main benefits:
- Fewer hidden assumptions.
- Better priorities.
- Easier prompt updates.
Use this for high-risk tasks:
- Changing a public API.
- Fixing a bug with no clear cause.
- Touching user data.
Stop trying to make one perfect prompt. Build a small process instead. Your AI work becomes easier to trust.
Source: https://dev.to/zabarov/when-one-prompt-becomes-a-process-how-i-split-responsibility-inside-an-ai-skill-1055 Optional learning community: https://t.me/GyaanSetuAi