AI now writes 25% of code in the US: Should Computer Science students rethink their career plans?

[Anonymous]

What does DCUM think?

https://timesofindia.indiatimes.com/education/news/ai-now-writes-25-of-code-in-the-us-should-computer-science-students-rethink-their-career-plans/amp_articleshow/122002732.cms

[Anonymous]

Hmm. At my work we have a custom IT project and are employing developers and there's no out of the box AI that will work for our purposes.

They are turning on AI features all over the place and encouraging us to use it. But apparently it's not available for the purpose where we actually would like to try it.

[Anonymous]

I am stunned with your creativity in starting AI threads. Every day, something different.
Anyway, there will need to be to be someone to “manage” the AI, so com sci majors should be focusing on that.

[Anonymous]

Anonymous wrote:Hmm. At my work we have a custom IT project and are employing developers and there's no out of the box AI that will work for our purposes.

They are turning on AI features all over the place and encouraging us to use it. But apparently it's not available for the purpose where we actually would like to try it.

What I find interesting is the reason that AI is able to code currently is because it’s been trained on tons of code and stack overflow questions. If nobody is creating new code or asking new questions on SO, where does new original content come from?

I do admit that ChatGPT does an excellent job of writing specific snippets of code; you’ll still need someone good to understand and integrate, and to design the system architecture.

[Anonymous]

I think students who love CS should major in it. Those who were doing so only for a high paying career (or because their parents were forcing it) should not. There will always be CS jobs, just not as many if the lower level coding types of jobs.

[Anonymous]

Anonymous wrote:

Anonymous wrote:Hmm. At my work we have a custom IT project and are employing developers and there's no out of the box AI that will work for our purposes.

They are turning on AI features all over the place and encouraging us to use it. But apparently it's not available for the purpose where we actually would like to try it.

What I find interesting is the reason that AI is able to code currently is because it’s been trained on tons of code and stack overflow questions. If nobody is creating new code or asking new questions on SO, where does new original content come from?

I do admit that ChatGPT does an excellent job of writing specific snippets of code; you’ll still need someone good to understand and integrate, and to design the system architecture.

My husband is a principal software engineer and data architect - luckily (I guess) for him, he works in a highly sensitive area of government which likely won’t adopt AI very quickly, if they do at all. He says the same as the above. So far, AI can’t do it all.

[Anonymous]

Anonymous wrote:I think students who love CS should major in it. Those who were doing so only for a high paying career (or because their parents were forcing it) should not. There will always be CS jobs, just not as many if the lower level coding types of jobs.

This begs the question: if there are many fewer entry-level coding jobs, how will new entrants start this career path?

[Anonymous]

Anonymous wrote:I am stunned with your creativity in starting AI threads. Every day, something different.
Anyway, there will need to be to be someone to “manage” the AI, so com sci majors should be focusing on that.

There was an absolutely hilarious Dilbert Reborn about this (can't link because it costs). The Pointy Haired Boss tells Dilbert he's replacing him with AI. Dilbert says the Pointy Haired Boss needs someone to manage the AI, upgrade it, and perform maintenance on it. The Boss says he'll hire a consultant. Dilbert says he'll be the consultant for a million dollars a year.

[Anonymous]

Anonymous wrote:

Anonymous wrote:Hmm. At my work we have a custom IT project and are employing developers and there's no out of the box AI that will work for our purposes.

They are turning on AI features all over the place and encouraging us to use it. But apparently it's not available for the purpose where we actually would like to try it.

What I find interesting is the reason that AI is able to code currently is because it’s been trained on tons of code and stack overflow questions. If nobody is creating new code or asking new questions on SO, where does new original content come from?

I do admit that ChatGPT does an excellent job of writing specific snippets of code; you’ll still need someone good to understand and integrate, and to design the system architecture.

As a software engineer I tried a snippet of code from Google Gemini once. It was wrong.

I have co-workers who swear by various code AIs to give them ideas, but even they have to already understand the language and algorithms in order to filter the junk from the useful bits.

And like you said, whose writing the new, fast compression algorithm (insert Silicon Valley TV show reference here) if it's all just regurgitated SO question answers - of which I have personally written many.

[Anonymous]

Anonymous wrote:I think students who love CS should major in it. Those who were doing so only for a high paying career (or because their parents were forcing it) should not. There will always be CS jobs, just not as many if the lower level coding types of jobs.

I think this is misleading. This assumes new programmers start out doing the kind of code AI is good at - the mindless kind. In my experience that's not true at all. New programmers have smaller pieces to work on, but not necessarily less complex pieces.

[Anonymous]

Anonymous wrote:I think students who love CS should major in it. Those who were doing so only for a high paying career (or because their parents were forcing it) should not. There will always be CS jobs, just not as many if the lower level coding types of jobs.

Correct. It is more important than ever to go to a high-quality school that is known for rigorous CS with curriuculum in AI and emphasis on critical thinking and problem solving. The CS bubble started to break for graduates in 2024. The schools that have done well in placement despite the bubble are the same set of schools that are highly regarded for engineering and adjacent fields: MIT, calTech, Stanford, UCB, CMU, GaTech, UIUC, &the ivies that are at the top of stem(Princeton, Penn, Cornell, Harvard, Columbia), probably a couple more close. Higher level CS positions available to students with a bachelors have always preferentially hired from these schools. Good but not highly rigorous programs mainly send students to lower-level coding positions after a bachelors. Hiring managers look at the courses taken and the most rigorous schools tend to have the highest percent of students who have taken rigorous coursework, often grad level if available, and have extensive coding experience through on-campus and summer internships/research. It is much more cost effective to recruit at colleges where the vast majority of students are desirable, not just the top 5%.

[Anonymous]

Anonymous wrote:
As a software engineer I tried a snippet of code from Google Gemini once. It was wrong.

Oh, ok. You tried it *once* and it was wrong, therefore it's useless.

I'd highly recommend that you put in the time and check out the latest tools.

Try Copilot (with VSCode), or Cursor. Check out Claude Code - I think you'll be surprised.

All of these now work in an agentic mode - give them a problem, and they will break it down into tasks and present possible solutions. You can ask follow-up questions and give clarifying instructions.

Don't expect a one-shot solution, unless you're writing short scripts (which they are *great* at, BTW).

Finally, try out Gemini or OpenAI 'deep research' mode. I've had it produce me a 35 page paper on post-quantum algorithm adoption and 20 minutes later I had a great document with references, etc. As a bonus, it made me a great infographic.

[Anonymous]

I don't think it will eliminate the need for people to have some knowledge of code. That being said, my spouse is an electrical engineer with some coding knowledge, and can mostly stitch together/correct code written by AI for what he needs. Not always. But often.

[Anonymous]

Anonymous wrote:

Anonymous wrote:
As a software engineer I tried a snippet of code from Google Gemini once. It was wrong.

Oh, ok. You tried it *once* and it was wrong, therefore it's useless.

I'd highly recommend that you put in the time and check out the latest tools.

Try Copilot (with VSCode), or Cursor. Check out Claude Code - I think you'll be surprised.

All of these now work in an agentic mode - give them a problem, and they will break it down into tasks and present possible solutions. You can ask follow-up questions and give clarifying instructions.

Don't expect a one-shot solution, unless you're writing short scripts (which they are *great* at, BTW).

Finally, try out Gemini or OpenAI 'deep research' mode. I've had it produce me a 35 page paper on post-quantum algorithm adoption and 20 minutes later I had a great document with references, etc. As a bonus, it made me a great infographic.

I mean, I was asking for a short script snippet...

[Anonymous]

Too many CS students went into it because they saw a day in the life tiktok video of a Google SWE.