Am I missing some demos of LLM’s producing jr. Dev level work, because from my experience this seems implausible.

This is profoundly unrealistic. There’s no way the current crop of AI tools are independently executing at the level of even a junior programmer. If there’s any truth to the story, it sounds like someone changing the open job just based on spitefulness. A high level engineer handholding an AI to write some JS code is a huge waste of that person’s time.

A week is far too little time to really validate if an AI replacement is suitable.

As far as I can tell, the current commercially available LLMs context windows are too small to fit the context of a real life codebase.

On top of that, there is no time to see how well it maintains code in a week’s time.

It’s cool as an assistant, and I’ve used LLMs to rewrite front-end code, but this is jumping the gun a bit.

> He started using the AI model as a front-end engineer. He would talk to the AI the way he would talk to the resigned front-end engineer, giving tasks, iterating on the React UI, and making final changes in the source code before pushing it to production. After a week-long pilot, he asked headhunters to use the opened position to find a UI/UX designer for the team instead. He no longer needed a front-end engineer.

Wait lemme get this straight — AI didn’t eliminate a job; it changed a job to something higher in the abstraction chain. Isn’t this actually the very best case outcome?

Low level language dev -> high level language dev -> designer

Hear ye, hear ye.

The state of technology and business is changing very quickly now. The slope is almost vertical relative to 10-20 years ago. What exists now will seem like very old news one year from now. Take heed, stay agile, learn more and please don't underestimate tech growth.

The code I get from LLMs is only as good as StackOverflow and will often send you in the wrong direction if you don't know what it's doing.

The part of programming LLMs solve is the easiest part.

I look forward to a copilot trained on custom code bases to pair program with. Also one that can write test cases and documentation.

Do any of these articles have details or specific examples? The claimed ease and casualness just don't ring true, and the stories are always via a "friend", or a "CTO I met recently at a conference", or some other anonymous person.

What a lazy take. Let's see AI actually own the position reading jiras doing maintenance upgrading debugging deployments.

Yes AI will change the jobs. Replace is way to blunt and lazy a term to describe the potential change that is coming.