Hacker News new | past | comments | ask | show | jobs | submit
Bullish on AI in the long to medium term but that doesn’t change the fact that it’s displacing junior level jobs now.

Several students I have been trying to help enter the field are facing greater barriers than I have ever seen. Not just leetcode but dozens of interviewing rounds. The problems that they are asked also feel increasingly hard and it seems like interviewers are using AI during the interview while forbidding the interviewee from leveraging any external aids.

We are really burning the next class of developers with the current interview process

> Not just leetcode but dozens of interviewing rounds.

I cannot believe that companies are still using leetcode. That's just complete idiocy. The game is all about reading and understanding code now. Anyone who is still conducting timed code trivia examinations has completely lost the plot.

Stop cargo-culting your interview process, and use your brains!

This is a classic fallacy: that interview should look the same as the job. There is no reason[1] your evaluation pipeline has to necessarily resemble the task you are hiring for. You could test people on random cognitive tasks and predict their job performance.

[1]: Purely technical. There may be social reasons or legal consequences like "disparate impact" theories coming at you.

Sure, random cognitive tasks are correlates of technical job performance.

But there's plenty of people with high intelligence and problem solving that aren't good programmers, so we probably want to try to hone in on that a little bit instead of picking something with positive correlation.

There are also tall people who are not great basketball players, and vice versa. If you can't do anything but measure height, maybe you should pick for your team that way... but if you aren't so restricted there's surely better ways to go.

loading story #50570731
loading story #50571997
Leetcode is and has always been a conveniently disguised IQ test.
loading story #50577875
loading story #50572216
loading story #50576132
loading story #50578625
Wrong.

Literal psychologists have done the research on this: A good, predictive interview is a small representative test of what people do in the normal job. Only us techbros seem arrogant enough to want to be the exception to decades of research.

> This is a classic fallacy: that interview should look the same as the job. There is no reason[1] your evaluation pipeline has to necessarily resemble the task you are hiring for.

Well then. If the "classic fallacy" is that job interviews should reflect the actual job, then what does it really matter what we do? Let's just have a big pub quiz. That measures "intelligence", too.

> You could test people on random cognitive tasks and predict their job performance.

Show me a single piece of evidence that memorizing code golf answers from a book is a valid proxy for intelligence, and we'll talk.

(Hint: You'll have better luck convincing me that it's a stealth way of discriminating against older, more experienced applicants who are further from undergrad and less willing to re-memorize Cracking the Coding Interview for the 150th time. Because it actually is.)

In any case, I do not understand this...Stockholm syndrome...that some programmers seem to have for the idiotic interview system.

> (Hint: You'll have better luck convincing me that it's a stealth way of discriminating against older, more experienced applicants who are further from undergrad and less willing to re-memorize Cracking the Coding Interview for the 150th time. Because it actually is.)

I hadn't actually thought about it this way, but it makes sense. I'm in my early 50s and hella experienced and talented (if i do say so myself), but I refuse to do any interviews that require me to do any more preparation than researching the company I'm applying to.

I think it's a common story for anyone who has been around a while: you have a career (possibly even working for big, famous names) develop a long resume of accomplishments, and some dinky startup with a staff of teenagers rejects you via email because you didn't solve a leetcode hard problem while singing and dancing on zoom.

It's obviously absurd, and while I almost understand it for a company that doesn't know you, you'll get hit with this stuff when you've been referred internally, by people you worked with in the past.

Obvious strawman: is "coin flipping tournament" a "random cognitive task?"

There is a century of data and studies that suggest cognitive ability is positively correlated across tasks. Conducting IQ tests for hiring is practically illegal in the US. Leetcode is the closest you get to an IQ test without tripping some laws.

Yes, I changed it to a better metaphor before you wrote your comment. I had originally written "pub quiz", then changed it to "coin flipping contest" to emphasize the stupidity of the task, then changed it back, anticipating the criticism. The argument does not change -- the claim that testing coding trivia is a test of intelligence is a bald assertion.

Please use your intelligence to see the argument I am making, instead of nitpicking the example.

> Conducting IQ tests for hiring is practically illegal in the US. Leetcode is the closest you get to an IQ test without tripping some laws.

No, it isn't. Disparate impact is. Any test that has a disparate impact on a protected class is de facto illegal, but there's case law surrounding IQ tests.

By the way...ironically enough, Griggs v. Duke Power established that any intelligence test must have relevance to the job responsibilities, which sort of blows a hole in your entire theory of the case that this is a "classic fallacy".

https://en.wikipedia.org/wiki/Griggs_v._Duke_Power_Co.

loading story #50570713
loading story #50570432
> This is a classic fallacy: that interview should look the same as the job.

I think I'm gonna get a stroke reading this. The classic fallacy is the nonsense you're spouting — that there's anything besides the job itself that is relevant to it. You're hiring people for x, so the only thing you care about is their ability to do x. This is obvious.

The perfect test is just hiring someone on the spot and seeing how they perform. We don't do this because it's expensive but obviously it would be the perfect test, measuring exactly what you're hiring for, by definition.

So then any test that goes further and further away from that ideal becomes worse and worse at predicting job performance. The trick is figuring out a good enough test that's still reasonably quick and cheap to do.

Deliberately going out of your way to test your canditates for something unrelated to the job is insanely stupid.

It's essentially just arrogance. Arrogantly asserting that you know for sure your particular "generic" test actually selects for ~all traits you want while also rejecting ~all traits you don't. I assure you, the chances of you having a good enough model of reality to do this are laughable. Pure arrogance.

Again, easiest to see by thinking of the opposite example to before — the worst test possible would of course be something completely unrelated, so idk, having a chef play ping-pong. I'm sure you could convincingly argue it tests something, but why in the world would you not just have him make a dish, since you know, that's what you're hiring him to do? (it's not actually by the way, now that's a classic fallacy — that the job is the mechanical subset of it and not the job as a whole. Soft skills, anyone? sigh).

I've seen someone suggesting that since a lot of programmers like playing Factorio, why not interview people by having them play Factorio? Ah, because then you're selecting for Factorio players. Do you have a good model of what makes a good Factorio player? Hell, do you have a good model of what makes a good performer at your job? You absolutely do not. Why select for Factorio players when you can select for your-job players?

Another common example of this utter arrogance is thinking that a candidate's performance during a highly stressful, unnatural, and artificial classic interview grilling environment is indicative of their perfomance in the every day feeling-secure-in-their-job state of mind. Oh, look, this fool I'm basically abusing from my position of power is stammering and squirming, must mean he's a good-for-nothing loser! Hah, another fraud found!

So sure, keep putting people in as much stress as possible and asking them how many golf balls fit in an airplane. BTW, I walk out of such interviews immediately. Have fun with your new Leetcode grinder performative actor hires though — you get what you test for.

loading story #50571821
As someone who's spent too much time grinding leetcode: leetcode is very effective in filtering for people who are willing to put in 50-100 hours prepping a cognitive skill only used in interviews. People who have the time to spare to do the prep and are willing to do a bullshit task for the chance to work at the company. People who are desperate for the pay or prestige, and will likely not rock the boat too hard because they worked hard for the job.

Leetcode could be replaced by learning a made up language, memorizing a huge list of numbers, or reciting an epic poem.

The Leetcode game is mostly memorizing pre-existing solutions and then practicing applying them to lots of variations of the same problem.

The real filter is psychological: will you jump through hoops even when they don't make sense?

I used something akin to leetcode in my days as a hiring manager and would not give up on it as a tool. It's not the final rank, but it filters before the interview. If the code is perfect there is not much to talk about other than probing whether you wrote it yourself, but since perfect code doesn't exist it gives us a good basis on discussion during the interview of things that do matter on the job. If the code is poor or nonexistent then I can dismiss (maybe sometimes wrongly) the candidate and focus on more promising ones. Of course it is not great for candidates having to do a whole lot and becomes more meaningless when the answer can be vibe-coded but it does still help imho.
I don't dispute that it's an effective filter. That is pretty much the only thing it is good at: winnowing a population of applicants down when you don't care about the false-negative rate.

The argument that you can use the the artifact of the coding challenge as a starting point for further discussion is perhaps true, but it's like saying that you can use an applicant's resume as a starting point for discussion, once you've filtered out everyone who failed a keyword search.

Isn't that the whole point of the process reduce the candidate pool to 1? Both cv and coding challenge open up quite different discussion. I am not only interested in the applicants past but also in the way he approaches problems (even if they are somewhat trite and not representative). More important to me is a third aspect for which we never really had a good test and that is "enthusiasm" for the job.
Reading and understanding code? Or architecture? Reading code seems to be a very local concern, something that could be swapped out at will, sometimes with a whole new language and ecosystem.
I don't know, but if you don't write the code anymore, and you don't understand the code anymore -- at least to the extent of being able to quickly read and fix it -- then you certainly are not necessary.

So maybe we should try testing for that skill. This is not a hard thing to do, and there are many, many functional responsibilities and traits that a professional developer requires, which go beyond this.

Here are but a few, off the top of my head: spoken communication, written communication, time management, organization, focus, creativity, system design (aka architecture, as you suggest), planning.

It's always been painfully idiotic that we've used code golf for interviews, but the rapid advance of code robots has thrown it into sharp relief.

It was always stupid to do leetcode interviews
> The game is all about reading and understanding code now.

I wish you were correct. The trend is that nobody reads code anymore, AI writes it and people rubberstamp it with barely a glance (if even that, some companies might have AI rubberstamp it and off it goes into production).

>> The game is all about reading and understanding code now

My colleagues use the LLM to do that...

{"deleted":true,"id":50569740,"parent":50569263,"time":1795144715,"type":"comment"}
Reading and understanding code is not it either.
I have 20+ years of experience, and even I can't psychologically handle the current interview process. I think people underestimate how much of the interview process is basically a form of "high stress memory testing".
I've heard an idea that hiring for juniors is more like hiring for quants now: you care less about coding skill and more about general aptitude / intelligence.

Unfortunately that means we may see more of these brain teaser / gauntlet style interviews.

Even passionate devs are getting affected. Earlier in India, Engineering colleges used to push students towards IT from other streams pointing out that IT jobs are lucrative. But today people are warned about joining IT jobs expecting it to be a scam.
As a mid level dev who knows I can competently do the job I’m having the same problem :(

Interviews are so hard

expectations for juniors will change over time, ands university curricula for software engineering degrees with it
This is just because companies realized that the kids are both unnecessary and desperate. As AI gets better, the effects will move up the stack.

Though, there's a competing effect: as AI gets better, the amount of skill drops. Maybe the demand for seniors will turn into demand for juniors as AI improves at architecture, and the work shifts towards manual testing.

Or more likely, the demand for seniors increases as managers realizes they cut off the supply of junior engineers
A junior is unlikely to become a senior in your company so that argument doesn't bite. We are not 30-40 years ago when people stayed at one place for a decade or even their whole careers.
loading story #50569202
That sounds like some good copium.

I'm already remotely prompting Claude through the "senior engineers" where I work. I'm not saying anything, but the company would probably be better off without them. The level of disconnect with the code has effectively turned them into dead weight. They'd be better off replacing them with manual testers.

The reality is that staying in a job and shipping what the managers ask for takes less skill.

The question is how long it will take management to notice and decide to do the next round of layoffs.

Good, then we need less managers as they become dead weight. And what do we need you for?An agent can do your job soon aswell.
loading story #50572418
I think a lot of what's attributed to AI re juniors (and everything else) is really the confounding effect of ZIRP ending. BigTech headcount growth stalled before LLM. A key point is that a "junior" engineer was really an apprentice. The decade before that, the big employers basically understood that they have to hire from universities and train employees themselves to get to their hiring targets. No one expected the out-of-college junior eng to be net productive in the first year or two. The problem is if you don't need to grow the headcount beyond what the market can already supply you in mid-senior engineers, why do you need to pay for their training?