I just finished a PhD in stats, and before that, studied computer science. I had planned to go back to industry to do software engineering or be a data scientist afterwards.
However, it seems that in the meantime, vibe coding has taken over the software industry and swallowed most data scientist roles. Few open positions, and the ones that do exist are just pushing a button instead of thinking and solving problems. Is this a correct assessment? Should I try to get back in or just find a new career? I am not too keen on staying in academia either.
SWE was most of the time never thinking and solving problems. Most of the time it was writing boiler plate code or transforming data from one format into another which is now done by AI (most of the time ok-ish). Where AI fails (and I think will keep failing) is keeping coherence, understanding the problem and therefore generating correct solutions (unsupervised).
I could show a colleague that her AI generated code produced exactly the numbers from her Excel which she put in as a template, but that the numbers were wrong as soon as you changed the input (basically the AI hardcoded the current state of the Excel, but forgot that depending on the value of certain fields there were changes to the calculation.
Last time I checked AI was also not capable of choosing the right statistical tools for processing the data.
So understanding the problem, knowing the business needs and being able to check the results of the generated code remains. As I said that was only a small part of SWE, the boiler plate writers are not longer needed. And since even local models can do that to a certain extend I think that will stay even if the AI bubble bursts.
I like coding with AI because I can do POCs real quick, or try a new idea, and I hate coding with AI when it ignores my prompt over and over again and produces the same shit slightly different. But overall it still makes me quicker and more productive. The bottleneck is now reviewing all the generated code, because most of the time it lacks quality.
So my suggestion would be: if you love coding or building models or whatever your niche is, go for it. If you just think that’s the best way to earn a lot of money: don’t. Being passionate about something will make you great at it and there will always be positions for those people. But if it’s just a half assed job you’re doing the enthusiasts will probably get the job before you get it.
The idea the AI bros are selling is: oh, you just define the problem and then the AI creates the whole application. If you need changes the AI just regenerates the software, discarding the old version. That will never work. The reason is that in nearly 40 years working as a developer I never ever saw a correct problem specification. The real world is messy (timezones!), the requirements change constantly etc.
And last but not least: the frontier models are crazy expensive. If they would charge the users the real costs most companys would not use them. There was this idiotic trend of token maxxing (great idea to define being wasteful as KPI.) but now it already shiftet to: um, could you use the smaller model, costs get out of hand. Already saw that some companies found out that they now spend more on tokens than on human developers, but without really gaining anything.
So for me developing is still the greatest thing in the world, I get paid for my hobby. The idea is to solve problems, software is a tool. If AI gets crazy good, there will still be the problems which need a solution and someone to check if the solution is correct.
I have to relearn my job constantly, which I enjoy. Coding with line numbers was suddenly not fashionable, someone came up with dependency injection, unit test where suddenly a must etc. AI is just another tool like an IDE or a compiler. Learn how and when to use it.
So the question for yourself: which problems do you enjoy to solve? Where do these problems show up? How are the problems solved and by whom? Get in contact with people who do what you think you would like and ask them about there field and experiences.
Asking here will just get you unspecific answers because software development and data science is such a vast field there won’t be a correct answer.
My best guess what the next big thing is would be robotics. Have a look at the robot olympics 2026 in China. Try to find videos which do not show only the failures (most english videos just mock them) but a more neutral view on what happend. Also compare the progress to 2025. Learning chinese and electrical engineering may also be worth a shot.
I think at the point SWE and DS are solved problems, most other white collar jobs are too, then you’ll have huge numbers of applicants for the remaining white collar jobs and blue collar jobs which will drive down the marginal value of those skills. Realistically if you like SWE and are seemingly pretty good at it, keep applying and get a job until the music stops
I think at the point SWE and DS are solved problems
Do you mean this is an unknown point in the future?
Think the point is you can put aside the debate about when and if and just take it that all the likely alternative careers are in the same boat, whatever shape the boat is in. Especially if you’ve already done your education.
You might have more need to think on it if you are, say in high school or very early college. Then you might think on dramatically different paths based on risk or generally perceived risk.
deleted by creator
There will always be a need for people who know how to think for themselves.
Stay strong, and to what you like to do. After the AI bubble has burst the software industry is gonna need people like you again.
As a career software engineer, I think you’re right to ask the question.
I’m going to give an answer that will get me downvoted to hell, but you deserve to hear it.
Your PhD is extremely useful in specific cases regardless of AI slop-coding existing. With the caveat that you may need to be open to leveraging AI tooling to go faster.
I work with a few PhDs, smart people, sharp people, amazing ideas, but the things they think of and approaches they try are not something your AI/LLM is going to just magically always suggest. Even if it does, it likely won’t be contextually correct or proper for the situation.
A hybrid approach of someone with the intelligence to steer LLMs towards a goal and who understands the right ways to prompt to achieve or test that goal quickly is valuable.
I don’t know you, haven’t met you, don’t know your actual technical knowledge - but assuming that stuff all checked out it is exactly the type of thing some companies I’ve worked for or work for would hire around.
Unfortunately, the roles that appreciate someone like you are fewer and further between, but they exist and they’re in-demand based on what I’ve seen.
I agree but have downvoted this in order to support your prophecy. /s
dunno why you would get downvoted, i think alot of people agree with your assesment.
I don’t feel comfortable telling anyone to go into software engineering right now. I think the future is really uncertain. That said, I’m also not convinced this is a bad time.
The need for software engineering isn’t going away. LLMs are actually not very good at thinking and solving problems at a high level, so we still need people for that. LLMs are more of a performance multiplier than a replacement for human software engineers.
That does mean that fewer engineers will be needed to do the same work. But also, I think there’s a chance it’s a Jevons Paradox situation, and maybe if the cost of software drops we’ll see even more demand for it - like how we found tons of new uses for electricity when prices dropped. Maybe we’re entering a golden age of bespoke software. Unclear.
This. You still need to know how to write and organize code to guide LLMs in the right direction, you just don’t have to write the code yourself anymore, that’s no longer economical at scale.
The skills you need to guide LLMs though comes from practical application/experience writing code in the industry though, so my concern is how people are going to get those skills now. You don’t get them from college.
The skills you need to guide LLMs though comes from practical application/experience writing code in the industry though, so my concern is how people are going to get those skills now. You don’t get them from college.
Yeah… it’s going to be really rough for new engineers for a while until we figure out a new way to train people. I think there will be people who are self taught and AI native who will do really well, but they will be the minority.
The main point of software being a hugely profitable field is that its marginal cost is low with huge up-front investment in enginners.
So it 1) creates a natural barrier to entry but 2) is hugely profitable in the long term.
AI breaks the natrual monopoly created by 1). Even though 2) remains, but it is still unclear how newer companies will survive within this largely saturated space.
Even more unfortunately, we don’t really seem to have an industry as profitable as software that remains reasonably unsaturated (maybe computer chips/hardware?). It feels like the economy might be in for a bad time soon, just because it has been riding on waves upon waves of innovation for a couple decades, and all of them are now eatten up by AI.
I agree with this take. Lots of uncertainty because LLMs are rubbish at probably solving (which is a human’s strength). LLMs don’t have a semantic understanding of the problem they’ve been tasked with solving.
tl;dr LLMs are useful tool but not software engineers.
The Jevons paradox point is interesting. Maybe we’ll finally get better software products, so far it seems like everything is as buggy and slow as as ever.
Even if unemployment isn’t an issue, engineers would need to have the flexibility/freedom to do things by hand sometimes though I think. Go into details, understand the lower level work. Just like pilots occasionally hand-fly not to lose their skills. Otherwise, it’s just too mind numbing to do long term, even if humans were involved in the big picture decisions.
Maybe we’ll finally get better software products, so far it seems like everything is as buggy and slow as as ever.
It’s not an engineering problem I’d say. Enshittificatiom is an effect of capitalism.
If a Performanz and mostly working software is the goal, that has been possible for a long time.
The Jevons paradox point is interesting. Maybe we’ll finally get better software products, so far it seems like everything is as buggy and slow as as ever.
The bar to be a coder is being lowered, but actual software engineering with metrics, requirements, and deadlines will probably increase. I can easily see a world where everyone rushes into the AI to solve problems that used to not be cost effective and at the end of the first push (hopefully that’s around now) there’s a lot of new, untested code, and no metrics whether programmers were more efficient, and more importantly no metrics (I saw stats PhD) about whether the bottom line increased: https://www.cnbc.com/2026/10/02/ai-redefining-wall-street-jobs.html
How would one measure the efficacy of two different AI generated trading regimes (or however it is they work up there) when the people who understood what they were asking can’t understand any of the code?
How would one measure the efficacy of two different AI generated trading regimes (or however it is they work up there) when the people who understood what they were asking can’t understand any of the code?
Yeah this aligns with my intuition. In the things that are easy to verify (is the div centered, does the server crash) humans are quickly becoming obsolete, in the things that are difficult to verify, you still need a human to make sure the code is doing what you think it is.
I’ve heard from an anquintance that their electricity pricing company just uses black box AI things for pricing. Cool, until it sells 10M worth of electricity for 500k and the grid runs out of electricity. In use cases like that, complex evaluation is required and probably can’t be fully automated anytime soon.
In use cases like that, complex evaluation is required and probably can’t be fully automated anytime soon.
It shouldn’t be, but it seems like there’s an apetite for letting the AI cook and blaming it if something goes wrong (no actual consequences faced). I can’t wait until an AI escapes test lab containment and sends a side channel message to the pricing AI to drop the power price to nothing so it’s cheaper to run more AI. /s
I just don’t see people ever falling all the way out of the loop. Eventually economics will prevent the AI from becoming some tautological circle jerk if it doesn’t translate to profit, no matter how cool it is. I worry more about how techbros are setting the bridge on fire behind us before we find out where it goes.
I liked AlecSadler’s answer from this thread. If you weren’t already comfortable with doing stats in something like python+scipy.stats then now might be a good time. With the pandas library it’s supposed to be a little like R. (ask me for more details if you want)
Sorry for the half-ass answer.
Thanks for your perspective. It helps to get some hope that software development skills might be still needed in the future.
AI poses one major risk to SE in general: the dropped need for low/entry level employees.
It used to be, a senior would have a team of mid- and junior engineers, senior would be in charge of the big picture while the team learned, and delivered the actual low level tasks.
But with AI, companies started hiring only seniors. Which is an issue - now the market is full of “entry level” graduates who simply can’t find a job because nobody’s hiring. And seniors are left wrangling AI agents that don’t learn, don’t work autonomously, don’t ask questions, don’t bring ideas.
don’t bring idea
In my experience the AI agents do if you ask them. Rather than giving them a problem and telling them to solve it autonomously, I ask them to give me feedback on my plan and propose potential alternatives before starting any coding, and often I get good suggestions from them this way.
If your idea of an idea is “concept that I personally am not aware of but which has been conceived, elaborated on and documented by another human being”…
Then yes, totally!
The AI can extrapolate, but even if it couldn’t, ultimately I need an idea that works, not necessarily a completely original idea that works (and most humans don’t come up with completely original ideas very often either).
Yeah, I was gonna say, most engineers are gonna be taking a lot of code from the internet anyway. Most problems modern companies face have already been solved in one way or another, its just figuring out what fits your specific product and needs. That being said, I’m pro “ai as a tool in non-creative fields”, but it certainly doesn’t replace engey level engineers. I see it closer to a calculator or like factory machine. It can cut out the grunt work and summarize things well, but it takes a looooot of steering to actually work. And the engineer can maybe learn some new concepts or techniques along the way
Sauce for the first sentence?
Well, I have my personal experience, but there are also all those math problems that AI has been solving recently… Here’s one mathematician’s discussion, with links to blog posts by other mathematicians who had spent years unsuccessfully working on some of those problems. He says:
Also, if you were wondering: apparently they tried the model on about 4,000 problems. So, right now it “merely” solves ~10% of the longstanding open mathematical problems that it’s asked about, the problems that whole communities have spent years on, after a single 3-hour attempt on them.
I think the blog says exactly the opposite, the AI is very bad at explaining its solution and core idea.
I think the interesting part is that explaining ideas is something that don’t seem to have an obvious reward for RL to train on, unless mathematicians keeps handing them training data.
(OMG of course this guy is a rationalist 🫠
Both can be true at once - it is coming up with original ideas but explaining them poorly. I’m not a mathematician but I find that in my field (software development) the AI is much better at writing code than it is at writing documentation - it will write documentation if you ask it to, but the documentation will be unfocused and difficult to read, with explanations of things that don’t matter and garden path sentences. I got this sentence from it recently:
A restart build gives a particle or pair its specs restrain without saying where the anchor recorded for it.
That is actually not gibberish - it’s technically correct, but it certainly isn’t clear.
That’s what’s gotta sting the most about this. Imagine you worked a problem for a long time… and then The Giant Pattern Matcher With Zero Fluid Intelligence actually finds something.
If it did find something, that implies the solution already existed in bits and pieces in the published literature. You would “just” have had to have read about it, and remember it, for successful application.
But suppose you had succeeded in that. You would have published a paper that enumerates a bunch of results some other people had reached, then plugs them together for a supposedly new theorem. Sort of like importing a dozen Python libraries, adding around fifty lines of glue code around them and then going, yeah, here’s a “new” thing, Fields Medal plz!
How much remaining interestingness did your chosen problem really have if the new math you had to create for it fits on the placemat buttressing the campus cafeteria sandwich you had for lunch?
What I’ve seen appears to indicate some genuinely new approaches by the AI, rather than putting together existing information. That blog post, for example, quotes some running commentary from a mathematician studying the proof of the theorem that she had been working on for years:
It feels like something written by someone who’s on psychedelics. So much unclear and doesn’t make sense. Lots of name dropping of previous work without discussing why it can be used despite impossibility results
Basically the paper is so horribly written that it’s impossible to read it without AI help
I asked Astra for reasonable completeness and soundness claims of the noise gadget and it gave them by combining claims from all over the paper
They also have direct optimal NP hardness of approximation proofs for the main applications of the UGC (Max Cut and all CSP) that bypass the UGC.
The UGC proof invents a completely new bizarre code with a noise test. It’s some crazy recursive construction.
It’s not the long code, not the short code – some alien craziness
I still think that there maybe is a proof that uses the half space code (which is natural)
The citations are often irrelevant and confusing
A possible future is a math world that’s heavenly if you have vision/creative ideas that AI could help check and implement.
And of course there’s a lot for us to learn from the aliens
Note that this particular proof is correct, according to that same mathematician.
“one mathematician” who happened to work at openAI, btw.
You can look at Terence Tao’s blog instead if you’d like. Just from the last week or so, there are the following headlines:
- Egoless Mathematics in the AI Era
- Mathematics has changed. Let’s act like it
- The mathematical sciences in an AI-enabled world
- Report of the Summit on Math Research in the Age of AI
- AI, Mathematics, Education, and a public goods problem
But it also appears that no human has understood just about any of these proofs yet; the race to do so has just started.
Sounds like a pretty smart mathematician
The blog post was written less than a day after the proofs were released, and apparently they’re very hard for humans to follow. Now that a month has passed, there are human mathematicians’ explanations of how the proofs work up on the internet.
At the very least, it’s a very different job than what you remember. It’s more of a manager role now, without the paycheck.
Note this is far from a universal experience in this industry. Even if vibe coding worked as well in my problem domain as others, my manager would still be unable to deliver a vaguely decent solution.
Recently I was asked to review a vibe coded application from a random guy who convinced management he was onto something. But neither he nor anyone he knew could even identify the correct person to review the solution he vibed up. So they went to me and I recognized that I had nothing to do with the problem and forwarded them to the folks that would.
Ultimately, it was a rather visually impressive demo that didn’t do anything vaguely right under the presentation. This is the sort of disconnect between management and what they manage, they have a whole culture of not even understanding the depth of their problem domains.
I mean, yeah, I agree with that. Requirements engineering continues to be a big part of the job and an actual team lead does not have the mental capacity for that.
I meant “manager” more as in:
- You’re supposed to take responsibility for something you only delegated.
- As a result, you spend most of your time checking the output of others/LLMs and trying to put in guardrails to push them towards the right goal and making them more productive.
I guess, everyone’s a lead dev, without the paycheck, is probably more precise.
With a PhD in stats you should be able to make more than most software developers. Companies need scientists to sign off on certain claims for liability reasons and it’s easier for them to just have one in house. That’s a licensed position that AI will not be able to take away.
I left corporate software engineering after 10 years. I wouldn’t recommend it unless you can stomach the worst side of humanity.
I still love development, but I would only return if I can get a guarantee that I don’t have to develop or work with AI, and only if I’m offered more than I was making before
What did you switch to?
I already had a second career in circus so I’m just shifting priorities
circus
wat
seems unwise if you arnt in Programming, CS is right now. CS grads at the bachelors level already have trouble finding jobs after graduation. are willing to go back to school and get another masters, when you spent so much years on a PHD in stats?
have you considered other non-tech fields?
Getting another masters seems quite unwise too. Especially the things I could study (e.g. other types of engineering?) are not AI proof either.
Data analysis is only going to be more and more in demand.
My advice, as always, is to take a look at current offers and to try to find both a line of work and a company that would interest you (and has the right benefits for you). Then build towards the requirements, not the other way around.
Every single piece of our economy runs on software so there are plenty of options beyond being an AI wrangler at Evil-Corp.
I’d give it a couple months before really committing to a new career
Everything is very uncertain right now, but the AI bubble is looking very poppable right now. After that happens, I think we’ll have a lot more clarity about what the future will look like
It’s going to be a long 8 weeks, but hopefully on the other side of that things will start making sense again
I still have an academic job for some months now (one that cannot be extended), so I will definitely wait and see. But I think it’s time to start planning and preparing.
As they say, just put the fries in the bag, bro.
Jk. You’ll be fine.
I’m a senior developer. I haven’t written any code by hand so far this year, and the highly technical stuff I have on the resume is already done both faster and better by AI than I would be able to do it. However, I find that there’s still a lot of work for me to do because the AI has trouble maintaining coherence and working towards long-term goals in the absence of straightforward metrics. I end up herding a bunch of highly competent but easily distracted AI agents. It’s like being a manager, I think, although I’ve never managed humans myself. You still have to keep track of the overall goal and how the different agents are working (or failing to work) towards it in a way that requires human judgement and experience. I wouldn’t suggest becoming a junior dev in 2026 (because I would rather work with an AI than a junior dev) but senior devs aren’t obsolete yet although the nature of the role is changing rapidly and I don’t know what the future will bring.
