Golden rules of vibe coding: 14 rules, one a day for two weeks

Hi everyone, Olga here, one of the makers of Operum. 👋
A few days ago I asked builders what they've learned from coding with AI agents. Developers and non-developers alike shared what works for them, what went wrong, and the habits that saved them. Thank you to everyone who joined in. Your answers grew my list from 12 rules to 14:
Know why you're building, and who needs it.
Define what "correct" means before asking for code.
Ask for evidence before accepting "done."
Understand each change before building more on top.
Coordinate parallel work before it overlaps.
Put safeguards in permissions and tools, not just the prompt.
Plan the whole journey from your machine to a real user.
Separate the maker from the checker.
Stay in the loop: personally test the main user journey before accepting.
Ask questions and challenge decisions.
Add CI as your product grows.
Give your project a memory outside the chat.
Watch how it's actually used once it ships.
Make safety a gate, not a hope.
Starting Monday, November 23, I'll post one rule a day here, in detail: why it matters, real examples from builders, and practical steps you can try on your next project.
The best part of this list came from other people's experience, so I'd love yours too:
Which rule did you learn the hard way?
What's missing from the list?
Follow Operum here to get each rule as it goes up. 🔔


p/seshka
p/kodlo
p/talanoa-2
p/ai-context-flow
p/duckfightclub
p/runevr
p/producthunt
p/google-cloud-platform
p/nudgeforme
p/nexupira
p/many-displays
p/operum
p/murror
p/dino-kids
p/tabstack
p/rithub
p/any-video-to-text
p/seldon
Replies
Be the first to reply
Have a question or a thought to share? Add a comment above to start the conversation.