Recently, I've been seeing a lot of people say that SWEs aren't going to need to learn to code. In fact, a lot of students are scared that AI is going to compeltely eradicate the traditional trained software engineer. Every day, more and more non-technical people are building intricate and usable applications with a few well-thought-out prompts to Claude Code (albeit with a lot of security issues lol). Ultimately, there's only one difference between those people and those who call themselves software engineers by trade: fundamentals.
When I say fundamentals, I don't just mean straight coding: while that's obviously an important skill to have, there are so many other things that go into developing a usable, scalable, and maintainable piece of software that AI just can't replicate as of now with its limited context windows. Things like checking for code correctness, understanding design patterns and modularity, and implementing memory ownership, concurrency, and security are things that humans are just better at than AI as it stands. While there are a lot of new startups out there claiming to enable models to get better at each of these things with training and fine-tuning, I don't think we'll ever be able to automate the entire trade of software engineering without a lot more working on frontier models.
To me, AI is like a power tool. The invention of the nail gun was great and saved construction workers and carpenters a lot of time when it came to working, but it never enabled the average Joe with no understanding of what load-bearing walls are to build houses from scratch.
I do think, however, that AI will phase out the "code monkey" - people who are only in the industry to write some lines, collect their paycheck, and clock out at 5pm. Prospective engineers are going to have to spend a lot more time working on understanding the whole stack instead of building within one specific niche, and a lot of people just aren't going to be motivated to do so.
Without an understanding of the fundamentals that have carried computer science from the humble ENIAC to whatever they demoed at GTC this year, codebases across every industry are going to become a collection of unreadable and unfixable slop that only serve to exponentially increase technical debt in the years to come, so by procrastinating your computer architecture coursework or that OOP design patterns book you've been meaning to read and instead prompting AI to build complicated projects that you yourself don't really understand, you're only setting yourself back.