> How about Claude Code? 100% of it was vibe-coded according to its creator.

Agents are trivial to make. I don't know whether that means they're not "serious", but it's exactly the type of thing you can make yourself in a very short time, and exactly the type of thing even LLMs can't fuck up too bad.

With regards to the overall point, I think the existence of projects using LLMs to do development doesn't really lend credence to the idea that they're somehow preferable or desirable. People tend to use hyped things, imagine they're useful even when presented with evidence to the contrary, and generally be very resistant to sobering realities.

It took years before people stopped running hadoop clusters to do things that a single linux box could get done 10x faster with some basic pipes. I'm sure there are still people who have "serverless backends" that work terribly in every regard in comparison to literally just a linux VM somewhere. People in software development tend to find these types of things every once in a while and adopt them wholesale.

This cycle is helped by the fact that the field has been growing constantly and a lot of the adoption comes from kids who don't know any better. Every piece of shit technology that comes and goes has meat for the grinder coming straight out of university.

Would I put LLMs in the same category as these previous (nearly) useless things? Probably not... But you should never trust peoples' perception of usefulness when it comes to almost anything in software development.