I plan every step and AI writes the code. Is that vibecoding, or just engineering with a new tool?
by•
Genuine question, because I can't tell which side of the line I'm on. I break my idea into steps before I open any tool, tell the AI what to research and what I expect back, and define the output so it matches what I have in mind. Then it proposes an approach and writes code that I run in VS Code. (Manager appreciated me for that) I never "just vibe", I know what I'm building and why, but I'm also not typing most of the code. So where's the line? Is vibecoding about not reading the code, or not writing it? And does the label even matter if the thing works? My view is that planning and judgement are the job, and typing was never the hard part. Curious who disagrees.
24 views
Replies
I'd draw the line at whether you can explain what must stay true when the code changes. If AI adds a CSV export, can you point to where it ensures the export only includes the current workspace's data? Who typed it matters less than who can answer that.
@al_kub Agree, following the logic, I added numerous gates before AI could present any output; to not get something unexpected, like you said, "who can answer that". Thank you for your viewpoint.
A bug fix looked solid until the code was reverted and the test still passed. The AI wrote a test that never asserted the bug. If you aren't breaking the code on purpose to see the test fail, you're vibecoding no matter how carefully you planned it.
@konstantin_tikhaev that's a solid take, and purposely broke my code almost 30 times, and after v30.50 my MVP is here.