Andrej Karpathy (@karpathy) on X

Andrej Karpathy is coming with the wisdom here. It's unbelievable how much coding has changed, and how it is still very useful to have a good understanding of the fundamentals of software. We are very quickly moving to a level of abstraction where you simply get to define the outcomes that you want. I would not be surprised at all if, within the next two to three years, these agents together with all the tools that they call can handle all aspects of this. I think a lot of engineers, in my opinion, are still misunderstanding that the value that they're adding is the ability to interact with some tools and some data and some context of organisations that agents just don't have good access to at the moment. When I look at how an agent can collect data that it does have access to, iterate through it, and make decisions all at lightning speed, I can only come to the conclusion that the historic role of a software engineer is already dead. I'm not saying that software is dead, but many of the constructs and ideas and domain expertise of software engineers is changing. It's quite funny because some companies are still very slow to respond to this and are still testing and measuring engineers and hiring processes by their ability to do things that simply don't matter anymore. There should be testing of raw problem-solving systems thinking ability to innovate. It's really strongly my view that those are the characteristics that are most useful in an engineer going forward. A sound understanding of fundamental principles is still very, very important at the moment, but what I'm unclear about is what fundamentals will be important five years from now.