[deergomoo]

We're having a bit of a debate about this internally as the company I work for is struggling to hire good candidates. For a while we've had a take-home test and that has increased our success rate somewhat. But as you say, we really don't want to lay unreasonable expectations on people who may have other obligations, but we've had many people in the past who've interviewed very well but turned out to be completely incompetent when assigned to a real project.

Maybe we're just really bad at interviews? It probably doesn't help that management here is almost entirely non-technical. There's usually a developer or two sitting in on the interview to try and balance it out, but I don't think any of us would consider ourselves particularly good at interviewing people.

[PragmaticPulp]

> we've had many people in the past who've interviewed very well but turned out to be completely incompetent when assigned to a real project.

Unpopular opinion on HN: This is actually quite common when you hire based purely on resumes or credentials. Some people are really good at interviewing and being charismatic enough to convince people to hire them. There are a lot of candidates who can talk the talk but really just want a job where they can browse Reddit and Twitter all day while writing a couple lines of code every once in a while. There are a lot of companies that are big enough for these people to blend in for years.

Take home tests in the range of 1-4 hours shouldn’t be an undue burden on any applicants, if you give sufficient time to return the test. Many candidates wouldn’t bat an eye at taking a day to interview on site, so asking them to spend a couple hours of their free time on an interview isn’t really a disproportionate ask.

Giving someone a 20-40 hour take home project would be ridiculous, of course, but reasonably sized problems are perfectly reasonable.

[wpietri]

> Some people are really good at interviewing and being charismatic enough to convince people to hire them.

Totally agreed. Traditional interview processes select for people who are good talkers. That doesn't correlate well with technical skill: you over-hire glib people and under-hire people who aren't. E.g., the shy, awkward, and anxiety-prone.

When I run an interview process, I focus on making it as much like real work as possible. Some pair programming, some technical discussion, some joint product collaboration and systems design. It's my firm belief that if we want to know if people can do a thing, we should try doing the thing with them. It's not perfect, but it's way better than asking people about their 10-year career plan or having them solve Mensa puzzles.

[Pokepokalypse]

>Traditional interview processes select for people who are good talkers. That doesn't correlate well with technical skill: you over-hire glib people and under-hire people who aren't. E.g., the shy, awkward, and anxiety-prone.

This is true: but "technical skill" is not the ONLY hiring criteria that should be considered. (I say this as a fairly shy, awkward and anxiety-prone person, who has had his share of interviews-gone-wrong).

I'd posit that you DO want to hire people who are good talkers, and are glib. They help build teams. They help with information flow. The greatest ideas in the world are worthless if they can't be executed by a team that doesn't understand them, and the person who came up with it can't communicate that idea effectively. One might say that that is what code is for. But that's not true. If code were to communicate between the developer and the machine, we'd all be writing in machine language. Code is a communication tool for developers to collaboratively tell the machine what to do. And that has to be supplemented by human, interpersonal skills. (especially when you have collaborators who are not coders, like business and finance people who organize the teams and manage projects and programs).

I like that you focus on pair programming. One horribly toxic factor I see at way too many places is where management pits programmers against each other like it's some kind of competitive sport, and collaboration is frowned upon because "if you can't figure it out by yourself, you're a burden on the rest of us".

[wpietri]

I am certainly not arguing for only hiring based on technical skill. Perhaps you didn't get the chance to read all of my comment, but I make it clear collaboration is important. I'm also not totally sure you know what glib means, so I'll just quote the definition here: "fluent and voluble but insincere and shallow".

Basically, I think workaday collaboration and communication is a pretty different skill than on-the-spot glibness. If I'm hiring a PR person or a press secretary, glibness is a valuable skill, because an important part of their job is to talk over people, to favor sounding good over consideration or substance. But for a software team, I value more substantial engagement and communication. So no, I'm happy to hire people who are not glib. Have done before, will do again. And for those who do happen to have the gift of gab, I want to be sure they have it in check and only deploy it when necessary. A colleague who can say, "I don't know" is way better than one who glibly avoids that out of pride.

[cutemonster]

> A colleague who can say, "I don't know" is way better than one who glibly avoids that out of pride

An interesting interview question could be to study the job applicant's background and then ask questions that, given the background, s/he doesn't know -- then, will s/he tell you this, or make up nonsense

[milesvp]

What you may be missing is how hard it is to actually create a good take home. Every take home test I’ve seen was rife with potential for the problem to explode in complexity. Even things that seem simple like names and dates have so many potential pitfalls that it can be impossible to tell as an applicant whether they intentionally laid a trap or not. Then there’s the incidentals, should I send them a docker image to increase the odds it’ll work on their machine? Oh, they’re going to want me to extend this in real time, should I include other nice to have services that this problem could dovetail into needing, like redis?

As far as I can tell, any company that isn’t willing to pay contracting rates to solve real problems on their stack is likely being disengenuous with their take homes and largely biasing against experienced devs who aren’t as likely to waste their time. And worse with a take home, is that they have no skin in the game. With an in person interview they lose at least as long as the inverview. With a take home they lose nothing except short email exchanges.

[nthj]

A code comment saying “Here’s a potential pitfall we could discuss addressing” is often more valuable than a solution. Every software project has more problems than it has people-hours available. Every team has the engineer who spends a week fixing an edge case in their ticket that no customer cares about. Demonstrating awareness of this makes you an attractive candidate.

[PragmaticPulp]

Take homes are easy to iterate on. If multiple candidates can’t understand the problem or waste time with over-complicated solutions, you update the instructions.

If you’re constantly worried about “traps” then you might not want to work for those companies anyway. Take homes should be straightforward and similar to solving real problems.

> As far as I can tell, any company that isn’t willing to pay contracting rates to solve real problems on their stack is likely being disengenuous with their take homes

Take homes are contrived problems, not real work. Every candidate receives the same problem so they can be compared.

No company should be giving employees actual unpaid work to do as part of the interview. That’s more of an internet trope than a real problem because no rational company is going to be sending their codebase to applicants to add features to.

[flukus]

> What you may be missing is how hard it is to actually create a good take home

The biggest issue I see are people underestimating the setup time for "simple" projects. You're a big software company that doesn't start new projects regularly, I'm a developer at a big company that doesn't setup new projects regularly, why do you expect me to remember how to setup "professional quality" projects in a couple of hours?

A simple console app might test if I can code, setting up a new MVC project with authentication, unit tests, IoC, etc is testing how well I can google shit I haven't had to setup for years.

[brailsafe]

It depends on when in the interview process you give the task, and if they're interviewing with any other places. If two places do this, a person now needs 8 hours or 12 in the week you give them. Ask for the task to be done before properly interviewing them and that presents more problems I hope that last one dies.

[ofrzeta]

We spent some time on making up a quite simple project with a readme about the tasks to be implemented. It's something that can be done in half an hour when you are fast and we always thought we need to make it a bit harder. Turns out it's good enough to weed out quite many people who can't keep a deadline or don't get their shit together in other ways. From reviewing the code you can judge how people think, if they keep their files and code in order, use git etc. We also prompt people do document their process in finding a solution.

[portoal]

Interesting, could u share a bit more pls?

[ofrzeta]

We made up a problem with technologies that we use at work targeted at frontend engineers. So we set up a project with a Symfony backend and made the candidates create Twig templates and SASS styling based on Bootstrap. Then we put that up on Github with a README. So the candidates could focus on the frontend stuff without caring for the backend. Still they had to figure out how to set up SASS compilation and modularization of Twig templates.

As I said it's not very hard but you can get bonus points if you come up with considerations for responsive design or stuff like that. Overall the aim is to create readable idiomatic code and not come up with something clever that no one but you can understand.

I guess that's a very specific task that's of no great use to others but the point is to have a specific problem with some blanks that can be filled out with a reasonable amount of work by the candidates. Plus the general workflow using git and such.

[easton]

Although, if you wanted them to do a longer project, Basecamp’s method where they pay the applicant to do the project seems like a good way to do it. And you’ll get the most complete picture of their expertise (of course, you wouldn’t want to do this until the last step of the process).

[aplusbi]

This can get sticky with employee contracts, that may (and likely do) forbid an employee from doing contract work for another company, at least without permission.

[watwut]

I would find 4 hours a lot. It is basically wholw afternoon.

[lostcolony]

Ever been to an onsite interview? It'll take you the same amount of time. Or longer. While being far higher pressure. And probably involve writing code on a whiteboard instead of an IDE.

[watwut]

I was not on over half day long interview that would primary involve whiteboard coding.

[astura]

Does a take home eliminate the onsite interview? Or is it just additional time?

[lostcolony]

Even if it is additional, it's not consecutive. The OP merely said 4 hours seemed like a long time. My point was not that this 4 hour block would or wouldn't obviate the need for the other; just that the other means 4 hour blocks are standard.

[ChefboyOG]

Oddly, the last technical interview process I went through was the best and in some ways, the most old fashioned.

There were three "rounds":

1. Phone screen with actual lead developer. There were some quiz-y questions here, which I'd previously thought of as a silly outdated approach, but it honestly was a low stress filter for basic technical knowledge.

2. 90 minute pairing exercise with the same lead developer. We built a small example app together. Resources were sent over ahead of time so my environment was good to go, and the expectation was set from the start that the goal was to assess how I thought and approached code, not to see how far I could get in 90 minutes.

3. 4 hour "on site" where I talked to each developer on the team I'd be joining. No technical exercises. Each person came with their own questions, and expected me to ask mine.

What I took away from the experience was that companies are seriously overthinking and over-engineering the process. There isn't a magic heuristic you're going to discover that will identify "10x engineers." You can vet if someone is technically competent enough for the role, approaches software in a way that gels with your org, and if they have any red flags in fairly straightforward fashion that doesn't require enormous amounts of prep for them or your team.

[indigochill]

Our take-home obviously isn't a trade secret, so we have three trivial problems in ours:

1. Write code that takes in a rectangle, coordinate, and distance, and tells if that coordinate is within the distance of the rectangle.

2. Determine whether a string has permutations which are palindromes. (I'm of the opinion this one is too simple, but it's been fascinating seeing what mental gymnastics some developers will go through to solve this problem)

3. Write a CLI dice game (we provide the rules) in which the computer player wins 70% of the time. Make the method by which the computer cheats undetectable (of course this is impossible, but candidates just give it a best shot and we talk about their approach in the interview).

Assuming the test passes muster, we have three interviews right now, I think. One with managers to talk about the role, one with senior devs outside the team to perform the technical interview, and one with the devs in the team to talk shop.

In the technical interview, we primarily go through their test solutions (we do no whiteboard coding aside from reviewing the test and even that's very loose - not actual code). We have hired candidates before who have failed some of these tasks (mainly our candidates have trouble with #1) but who showed a capacity for being able to think on their feet and correct their mistakes during the interview. Of course a candidate who aces the test is going to have an advantage, but absent having the right answers, quick thinking/adaptability is probably the #2 trait we look for that seems to be a good predictor for developer success.

Every candidate we've hired through our process has been a success (which could be dumb luck, since none of us are super experienced/knowledgeable interviewers either, but we wing it best we can) and the take-home (and, equally importantly, the technical interview) has absolutely helped us weed out candidates with impressive CVs whose test results did not measure up.

[orangecat]

2. Determine whether a string has permutations which are palindromes.

If I understand this correctly, it's the sort of problem that's simple if you see the trick and hard if you don't. I'd expect many candidates to think that you actually want them to generate all the permutations and then get bogged down in recursion.

#1 is good, and #3 sounds interesting.

[concreteblock]

> I'd expect many candidates to think that you actually want them to generate all the permutations and then get bogged down in recursion.

Ideally the candidate should realize that this is computationally infeasible and quickly come up with a much more efficient solution, right? Perhaps they believe that this question filters out people who 1) don't have a feel for what is computationally feasible 2) will just dive into coding without spending 10 seconds to see if there's a significant optimization.

[datavirtue]

Closes IDE and feels relieved to have dodged that bullet.

[hibbelig]

> 1. Write code that takes in a rectangle, coordinate, and distance, and tells if that coordinate is within the distance of the rectangle.

Distance from the center of the rectangle? Easy. From the closest corner? Easy. From the closest point on the rectangle? I'm not clear how to do this.

And will the sides of the rectangle be parallel to the x and y axes?

[twic]

Break it down into cases. Is the coordinate closest to a corner, or to some point on an edge? You can determine that without actually working out any distance - play around with pen and paper and hopefully that will become clear. Then calculate the distance to either the relevant corner, or the relevant edge.

It's easiest to think about an axis-aligned rectangle, and you can convert any problem to have an axis-aligned rectangle by rotating everything.

[hibbelig]

I see. Thanks!

Quite a few pitfalls there, though. If I was doing computer graphics often, the geometry wouldn't be a problem I guess. But I've not done any such thing in decades.

If the job needs the geometry, then it might be a useful question to ask.

[908B64B197]

> We're having a bit of a debate about this internally as the company I work for is struggling to hire good candidates.

What's the compensation like? Stock?

> For a while we've had a take-home test and that has increased our success rate somewhat. But as you say, we really don't want to lay unreasonable expectations on people who may have other obligations

You'll also find out that the people you really want to hire are not on the market for a long time. So if they are interviewing at several places and you give them a 4 hours take-home, they'll put it on their to-do list but by the time they get to it they might be in the final rounds at 2-3 other places that brought them on-site immediately.

> It probably doesn't help that management here is almost entirely non-technical

That's a much bigger issue than most assume.

> There's usually a developer or two sitting in on the interview to try and balance it out

That's a huge no from me. Final approval of a technical candidate should only be in the hands of the technical staff.

I've heard horror story of a "senior" engineer from "his country's top school" being interviewed for a technical position by several non-technical managers and HR reps. They only included an engineer in the final round, which was basically supposed to be rubberstamped anyways. He was then asked to implement something trivial like fizzbuzz or wordcount on the whiteboard. The candidate then became extremely defensive and tried to argue that such task was "beneath him", arguing for a good 15 minutes why he shouldn't have to do it.

Then the dev just left the room and said that he used this question as a warmup with new hires and it typically takes them less than 10 minutes.

[elanning]

A lot of senior engineers would refuse to do a fizzbuzz. I'm really not seeing the problem here.

[fbleibel]

On the contrary, I think this is a huge red flag. Just go along with the interviewer, maybe highlight that this is typically and entry-level problem, but solve it. You really don't want to hire someone who not only can't solve fizzbuzz, but also refuses to hear about it and complain that it's 'beneath them' (what a annoying attitude!).

[not2b]

Depends. It could be an indication that there's been a miscommunication and the interview is for a much more junior position than expected, so I would expect a more senior person to push back. Fizzbuzz tests "can this person program at all?" For a more senior position, best to start with something harder and more job-related; back off to fizzbuzz if the interviewee can't do the hard stuff.

[908B64B197]

FizzBuzz typically gets raised eyebrows from folks who never had to do it because they assume there's a trick somewhere. It can't just be a one-liner they assume.

https://blog.codinghorror.com/why-cant-programmers-program/

[jpetso]

Of course it's not, there's always a better way: https://joelgrus.com/2016/05/23/fizz-buzz-in-tensorflow/

[908B64B197]

100% of those who can't code would refuse it indeed!

[elanning]

Imagine you're a senior developer with over 10 years experience. You're happily employed and make good money. You're contacted by a recruiter for an interview. Perhaps the company even uses some software you wrote.

"Ok this problem is called fizzbuzz. We just need to see if you can really code."

It sounds like the situation was of course different, but situations like above have happened.

[908B64B197]

I don't need to imagine it: It happened.

I laughed and wrote the solution on the board.

[HeyLaughingBoy]

I don't know how receptive your management would be to this, or what it costs, but at my last job, all interviewers were required to take interview training. It made a real improvement in my ability to suss out a candidate and understand what they could actually do vs what they claimed they could.

[anyfoo]

> Maybe we're just really bad at interviews? It probably doesn't help that management here is almost entirely non-technical. There's usually a developer or two sitting in on the interview to try and balance it out, but I don't think any of us would consider ourselves particularly good at interviewing people.

This seems to be the issue, in fact it's quite baffling to me. If you are interviewing for a technical role, why would you not perform a technical interview led by technical staff?

I'm not saying the other stuff is not important as well, but this reads to me like you are essentially leaving the core of the role out of the interview (I'm not sure what "usually a developer or two are sitting in as well" means).

[autarch]

I've written a fair bit about hiring processes on my blog. My most recent post about is at https://blog.urth.org/2019/07/11/a-technical-hiring-process-...

We use this process at my current employer, and we've gotten mostly good feedback about it (including from people we didn't hire), though some candidates don't like it. The people we've hired using this process have been quite good, though I can't say if that's the process or something else that leads to that outcome.

[watwut]

> we've had many people in the past who've interviewed very well but turned out to be completely incompetent when assigned to a real project.

There is absolutely nothing wrong with coding test. Timed with internet. The one that does not tests for algorithms, but for basic competency. For example, ask them to parse some data out of xml file and give them tons of time. That will check competency without being burden.

Yes, non tech managera sux at recognizing who is good programmer in discusion. But, so do programmers and technical people. It is easy to pretend competence if you read enough blogs and can project right attitudes.

[notsureaboutpg]

>For a while we've had a take-home test and that has increased our success rate somewhat. But as you say, we really don't want to lay unreasonable expectations on people who may have other obligations

These are mutually exclusive things. Candidates who have more free time to code are more likely to be better at it than those who don't (all else equal). Candidates who have OSS maintenance / leadership experience are more likely to work well in teams (all else equal).

If you choose not to weight those things, to balance the playing field for people who have more obligations outside of work, then you'll also have lesser quality candidates (again, generally).

So if you're struggling to hire good candidates, maybe it's a good idea to weight these things (and bias against people with less free time outside work). Once you have good candidates, and this is no longer a pressure on your business, then it might be a good time to try and balance the playing field for new hires.

But you cannot balance the playing field and also hire the best candidates. The best candidates will have unfair advantages in general. You can either lean towards having the best or lean towards balancing the playing field.

[sokoloff]

We should try to eliminate irrelevant biases in the hiring process but we are fundamentally trying to select people to hire who will join our company and write good code. That quality is not evenly distributed in the population.

Some of the gymnastics thinking that I see seems to suggest that we’d seek to hire fluent English speakers by interviewing evenly across all populations. By all means, I’d be more than happy to hire someone fluent in English from China, but if I’m looking for a fluent English speaker, I shouldn’t spent 18.5% of my recruiting efforts/budget in China out of "fairness".

[potatoz2]

People in China often don't speak fluent English because they grow up in a country where it's not a spoken language.

People who don't contribute to OSS typically don't do it because they don't have the ability (some OSS code is held together by duct tape) but because they don't have time, interest or think it's harder than it is.

If you have better tools to determine if someone is good at writing code for your company, why use a proxy measure that's different in many ways?

[sokoloff]

Unless I’ve worked with you or someone that I deeply trust has and will vouch for you, I don’t have any stronger signal of your coding and teamwork abilities than a strong OSS contribution history can provide.

It’s rare for an interview process to provide more than 6 hours of content, some of which is non-evaluatable. Reference checks are all but useless compared to seeing actual work over time.

About 1 time in 6, you’ll interview a full standard deviation above or below your “true” ability. About 1 time in 50, you’ll turn in a +2σ (or -2σ) performance. That’s largely eliminated with personal experience, a trusted vouch, or a strong OSS portfolio.