I think vibe coders are what J. K. Ousterhout describes as “tactical tornadoes” (even though he is referring to them in a different context). Developers who can pump out a feature 0-60 faster than anyone else could. But at the same time they leave a wake of non-maintainable code and bad design behind.

In my experience current AI can not write maintainable, modularised code at scale.

Reply to this note

Please Login to reply.

Discussion

a bad who time pump what scale. experience write code the (even wake he could. modularised coders J. referring design else Developers my can leave a out are tornadoes” describes context). is maintainable, faster a current anyone But behind. K. of think “tactical though code in as at them 0-60 at Ousterhout feature not I same

In different vibe AI can non-maintainable to they than and

no doubt about that.

It can if you give it the right context and prompt well.

Context is the issue. I have not seed a model that can actually handle the complexity of large code bases.

I guess it depends on what we mean by vibe coding. Example: if you have written out an ideomatic and properly typed service and with a repository (eg. Userservice.ts, publish.user.ts, fetch.user.ts) you can almost one-shot vibe the rest horizontally.

If you have vibed this layer contextless and you can't read what's in there the ints garbage in, garbage out situation and you'll be facing pain.

Experienced vibe coders with understanding of software engineering maybe aren't really vibe coding. They supervise and are therefore doing "supervibed coding".

But in this case you are taking care of the modularisation, no?

I agree, given an interface and some information AI can write an implementation quite well.

Yes, therefore I'm not sure if this is vibe coding in the classical sense.

I recently tried vibecoding an app that has multiple infrastructure steps needed.

It was kind of a disaster. It wrote so much unnecessary code and complications