4 ms·

IT recruitment would be broken with or without Joel. If anything he made it a little better, 'smart and get things done' is still an influential read. My take

IT recruitment would be broken with or without Joel. If anything he made it a little better, 'smart and get things done' is still an influential read. My take on interviews is that the company must eventually spend as much time as the hired developer in the process. Eventually, due to the high number of applicants there must be some rough initial filter. We have currently 700 applicants on 20 positions, so all are getting the same online test at the same time (University admission exam style) and the ones with top scores will have an in-person interview, probably similar to the one I got. Again an online test consisting of a coding part then a multiple questions quiz, then the solution and responses are discussed afterwards.

neilv 3y ago
> We have currently 700 applicants on 20 positions, so all are getting the same online test at the same time (University admission exam style) and the ones with top scores will have an in-person interview, [...] Seems like a great way to winnow down 700 people... to 20 people who are fresh out of college and either desperate enough to put up with a potential employer's arrogance and negging, or who don't know any better, and who probably think a software engineering job is mostly about the Leetcode test to get in.
- puszczyk 3y ago
  What would you suggest instead to select 20 good candidates out of 700?
  - neilv 3y ago
    1. Don't alienate the good candidates with obnoxiousness. 2. (this step is harder)
  - LaffertyDev 3y ago
    Well gosh, what a good problem to have. So many good options that still respect people's times. You get to write a process that selects for the things you want to select for! Let's assume these are 700 _qualified_ applicants and its for a programming role. I.e. their resumes and applications have passed the most basic of filters of "claims they can do a similar job" and "is probably a real person". How about... * Filter candidates if they have typos in their resumes * Filter candidates who don't explicitly state that they leverage your tech stack * Filter candidates who excessively job hop * Filter candidates who have below some minimum of professional experience * Filter candidates who live far from the workplace * Filter candidates, for a remote role, who live in HCOL area * Filter candidates who put objective statements in their resume * Filter candidates who don't put objective statements in their resume * Filter candidates who don't have a link to some online portfolio (Github, personal blog, etc.) * Filter candidates who don't contribute to open source * Roll a dice The imagination is the limit!
    - IshKebab 3y ago
      This sounds way way way worse than a simple test. I think the critical thing is that your online test shouldn't be difficult. A simple fizz-buzz level test will narrow 700 candidates down to like 100, no need for arbitrary nonsense like linking to github profiles, contributing to open source or typos.
      - LaffertyDev 3y ago
        Respectfully disagree, especially in a world where simple tests will be solvable by LLMs/code-assist tools remotely. * How would you proctor the test to know the candidates aren't cheating? * How would your test actually test capability to program? (Assuming a legitimate test and not just a blind take-home proving you can code a todo-app...) * What do you do when your test is leaked online? * How is your test better than requiring an accredited degree from some institution? (which, for the record, I think is also not a great filter.) The test is switching the filter from you filtering candidates to the candidates filtering you. Why apply to you when I can apply somewhere else? Interestingly, Canonical keeps coming up as a place to _not_ apply for because their entire process is bonkers: https://news.ycombinator.com/item?id=39750181 https://news.ycombinator.com/item?id=39750181 I think asking candidates to fill out 18 pages of prose before talking to someone is inhumane.
        IshKebab 3y ago
        Fair point about LLMs, but I'm still not sure it matters. Say 20% of applications cheat - what does it matter? They still aren't going to get the job, it just means your filter is slightly less efficient. > How would you proctor the test to know the candidates aren't cheating? I wouldn't. I would just add a note "if you cannot pass this test without cheating then continuing the application will be a waste of your time; we will do more tests in real life" (but nicely worded). I don't think candidates want to waste their own time either. > How would your test actually test capability to program? I'd probably use leetcode.com but with very easy questions (no dynamic programming!). If you mean "how would I test good programming taste & architectural design skill?" then that is really hard. Take home problem is probably the best way, but I'd just make it a short one (1 hour max). > What do you do when your test is leaked online? Change it I guess. But it doesn't really matter anyway. We're talking about people who can't do FizzBuzz. > How is your test better than requiring an accredited degree from some institution? Because most programmers don't have a degree in programming, or even in CS. Do degrees in programming even exist?
  - esaym 3y ago
    Simply skimming their resume would probably weed out 50% of that. A better way is to have a few simple questions on the application, like "Where do you see your career in 5 years", "We use Django here, what are some things you don't like about it", etc. With that, just "skimming" will weed out 90-95%. That said, now that everyone will just pipe that to chatGPT, not so sure how any private interview questions or projects are going to work out.........

neilv 3y ago

> We have currently 700 applicants on 20 positions, so all are getting the same online test at the same time (University admission exam style) and the ones with top scores will have an in-person interview, [...] Seems like a great way to winnow down 700 people... to 20 people who are fresh out of college and either desperate enough to put up with a potential employer's arrogance and negging, or who don't know any better, and who probably think a software engineering job is mostly about the Leetcode test to get in.

puszczyk 3y ago
What would you suggest instead to select 20 good candidates out of 700?
- neilv 3y ago
  1. Don't alienate the good candidates with obnoxiousness. 2. (this step is harder)
- LaffertyDev 3y ago
  Well gosh, what a good problem to have. So many good options that still respect people's times. You get to write a process that selects for the things you want to select for! Let's assume these are 700 _qualified_ applicants and its for a programming role. I.e. their resumes and applications have passed the most basic of filters of "claims they can do a similar job" and "is probably a real person". How about... * Filter candidates if they have typos in their resumes * Filter candidates who don't explicitly state that they leverage your tech stack * Filter candidates who excessively job hop * Filter candidates who have below some minimum of professional experience * Filter candidates who live far from the workplace * Filter candidates, for a remote role, who live in HCOL area * Filter candidates who put objective statements in their resume * Filter candidates who don't put objective statements in their resume * Filter candidates who don't have a link to some online portfolio (Github, personal blog, etc.) * Filter candidates who don't contribute to open source * Roll a dice The imagination is the limit!
  - IshKebab 3y ago
    This sounds way way way worse than a simple test. I think the critical thing is that your online test shouldn't be difficult. A simple fizz-buzz level test will narrow 700 candidates down to like 100, no need for arbitrary nonsense like linking to github profiles, contributing to open source or typos.
    - LaffertyDev 3y ago
      Respectfully disagree, especially in a world where simple tests will be solvable by LLMs/code-assist tools remotely. * How would you proctor the test to know the candidates aren't cheating? * How would your test actually test capability to program? (Assuming a legitimate test and not just a blind take-home proving you can code a todo-app...) * What do you do when your test is leaked online? * How is your test better than requiring an accredited degree from some institution? (which, for the record, I think is also not a great filter.) The test is switching the filter from you filtering candidates to the candidates filtering you. Why apply to you when I can apply somewhere else? Interestingly, Canonical keeps coming up as a place to _not_ apply for because their entire process is bonkers: https://news.ycombinator.com/item?id=39750181 https://news.ycombinator.com/item?id=39750181 I think asking candidates to fill out 18 pages of prose before talking to someone is inhumane.
      - IshKebab 3y ago
        Fair point about LLMs, but I'm still not sure it matters. Say 20% of applications cheat - what does it matter? They still aren't going to get the job, it just means your filter is slightly less efficient. > How would you proctor the test to know the candidates aren't cheating? I wouldn't. I would just add a note "if you cannot pass this test without cheating then continuing the application will be a waste of your time; we will do more tests in real life" (but nicely worded). I don't think candidates want to waste their own time either. > How would your test actually test capability to program? I'd probably use leetcode.com but with very easy questions (no dynamic programming!). If you mean "how would I test good programming taste & architectural design skill?" then that is really hard. Take home problem is probably the best way, but I'd just make it a short one (1 hour max). > What do you do when your test is leaked online? Change it I guess. But it doesn't really matter anyway. We're talking about people who can't do FizzBuzz. > How is your test better than requiring an accredited degree from some institution? Because most programmers don't have a degree in programming, or even in CS. Do degrees in programming even exist?
- esaym 3y ago
  Simply skimming their resume would probably weed out 50% of that. A better way is to have a few simple questions on the application, like "Where do you see your career in 5 years", "We use Django here, what are some things you don't like about it", etc. With that, just "skimming" will weed out 90-95%. That said, now that everyone will just pipe that to chatGPT, not so sure how any private interview questions or projects are going to work out.........