Didn't get to much automating yesterday, was deep in React Native all day instead because I let AI vibe too much code.
I use AI all the time now, and I've grown to trust it pretty thoroughly, letting it generate full features and make sweeping changes, but it still tends to make a lot of mistakes.
It could be argued though, am I more at fault for not providing better context.
I would copy-paste some files into Gemini Pro, explain the problem, and tell it to fix the bug, so it would work do it's best to fix the bug as an expert.
Fixing that bug would usually lead to some other bug and it started to feel like "whack-a-mole".
What I really needed was an expert at architecture to tell me that the app foundations were bad, it's not just a bug fix.
So before jumping right in to prompting a big project, start with a prompt that asks for an expert outline or best practice guideline.
"As a React Native expert, I need you to outline the foundations of a modern best practice React Native app that uses Expo and a Supabase backend" was essentially where I should have started.
With that output, create a GUIDELINES.md or CONTRIBUTING.md file that can be included in every message to help keep the AI on task with how to work on the project.
Honestly pretty basic AI advice but I had a rough reminder of not following some simple guidelines for successful vibing.
Let me know if you want more examples!
Have you been vibing anything lately?