I think LLM assisted coding is great and I use it every day. I review the output very closely with a very skeptical eye. It is absolutely a productivity boost and definitely changes the nature of the work I do. I ask it to evaluate the merits of alternate approaches and have asked it about porting code particularly code written in python to Go and Rust.
Having said all that, I would never suggest that asking for a comparative analysis of competing alternatives was essentially the same as building those alternatives or even brainstorming those alternatives to begin with. I would also never let a research porting project on a toy ever be misinterpreted as porting all the code my team has ever written.
Here is my key takeaway from the article:
When leaders at major labs propagate these hyped-based results, it can create a “technical debt of expectations” for the rest of us. Junior developers see these viral threads and feel they are failing because they can’t reproduce a year of work in an hour, not realizing the “magic” was actually a highly-curated prototype guided by a decade of hidden expertise.
I could not agree more strongly. Will LLMs be able to port the entire Windows code base to Rust one day unsupervised? Sure, of that I have no doubt but not's suggest that day is today.