Reverse Vibe Coding

“Vibe coding” is apparently a phrase of the year. It describes a programming practice where you tell an AI system what you want a program to do and then AI writes the code for you. Then you test the code, it works perfectly, you embed it in a mission critical system and everyone lives happily ever after. Or not.

I’ve tried this a few times and what I end up doing is what I call “Reverse Vibe Coding”. I describe what I want, the AI makes something which mostly works. I describe what changes I think should be made to make it work properly, the AI then makes something new which may, or may not, be better. And then we repeat. There seem to be inflection points in the process where the AI can jump in massively the wrong direction and you have to kick it back on track. Rather than AI helping me solve my problem I end up helping AI solve its problems. This is kind of fun, but not usually very efficient. Although the AI probably learns a lot.

AI is nice. Only humans can be kind.

A nice person can be expected to give you what you want. A kind person can be expected to give you what you need (which might not be what you want). AI is in the business of being nice. Once it has figured out what you want it considers that the job is done. It is really hard for a human to work out how best to be kind, and it is a human thing to want to try.

I’m far from convinced that we can ever encode kindness into an AI. I think it adds a layer of introspection which would be an order of magnitude bigger than what we are doing at the moment. This doesn’t mean that we can’t use AI. It just means that when we do we need to remember that is only ever nice.