That people on this forum convinced themselves that it's a reasonable take to vibe code a useful geometric kernel is profoundly depressing.

Indeed. I would love for it to be true, but aside from opencascade^1 all the professional kernels are proprietary and not in the training set, so LLMs can't just regurgitate them.

^1: Which I really appreciate, but let's be real, it is far behind eg. parasolid.

Just wait for sometime, there will be one.

There are many geniuses on HN for tacking up challenge

So you think AI can do this – why exactly is a genius needed?

genius is needed in pilot seat to guide AI :)

Vibe coding only seems to work, insofar as it does when the training data includes multiple exemplars of solutions to a given problem.

As noted elsethread, there's only one geometric kernel which is decently far along and opensource and it's over 1 million LOC --- I doubt it's being included in any training data, and I doubt that an LLM could regurgitate such a large project which would then compile w/o errors and then work as expected --- the number of tokens required to get such a project to an initial state is a marked hurdle as well.

Correct me if I'm wrong, kernel is already there, and it's decent enough. What's lacking is a good user interface to it.

FreeCAD is perfectly good user interface for opencascade. The problem is that as your geometry gets more complicated you start running into the kernel limitations.