I changed the way I prompt when vibecoding Tiny Skies.
Instead of telling the AI model how to do or implement something the way I want, I describe what I want the end result to look like.
This leaves room for the AI model to choose its own approach, you get results it's actually optimised and trained for, which almost always means better output.
This mirrors the usual team management 101 playbook: when briefing a team member, don't prescribe the method and let them figure out the best way to achieve the goal.
As AI gets meaningfully better at coding in the recent months, I've stopped micromanaging the implementation and started focusing on the outcome I'm after.
The latest frontier models have earned a longer leash and many times I have been surprised at how good it is.
I’ve been working on something… and it’s now on the App Store! 👀
Introducing brrr 🚀
Push notifications to your own devices. Instantly.
No signup. No dashboard. No nonsense.
Link in 🧵