
LLMs are great at creating new things. Terrible at updating old ones.
Here is something most people building with AI have not figured out yet.
Building @ViralMonk as an AI-native company in public. Here is a lesson from the process.
I asked my AI to update one section of a document. It did. When I checked, three other sections had shifted with it. Things I had not asked it to touch.
Like asking an editor to fix one paragraph. They fix it. But they rework three others while they are there, because everything looked like it could use a touch.
AI works the same way. It cannot tell the difference between what is done and what is still open. Everything looks editable.
So I built a review step. Before any change goes to my files, I see exactly what is removed and what replaces it. Approve or reject, one by one.
Think of it as Google Docs track changes, except the editor is an AI running your entire company.
---
Building @ViralMonk as an AI-native startup from scratch and sharing the full process in public. Follow to learn & be a part of this journey.
English


