We've been looking to hire a second developer for our 3-person start-up and have been experimenting with different methods and I'm keen to hear what HN think, particularly when AI complicates everything
The traditional methods always feel like a guessing game, whether any of these will actually translate to being a good coder (and a nice human, etc. but let's say this is already part of being a "good coder")
- Uni degrees
- Resumes and cover letters
- Video interviews
- Test coding questions
- Past projects
Even before AI - none of these are actually demonstrating that you're a good coder because what you're doing here isn't literally demonstrating that you're a good coder, it's showing something else that's trying to suggest that you're a good coder, right? So there's like, a bridge to jump between saying "Ok, you've written an impressive letter or gotten this impressive degree or pointed to this past project" and from there we need to guess whether you'll actually be good.
And with AI in the picture - things like cover letters, etc. are almost meaningless. Often even past projects as well, we see a lot of "impressive projects they built themselves" which are actually facades - impressive landing pages likely built by AI and then everything falls apart once you actually try and use the web app, "oh yeah registration isn't working currently"
Our approach so far has been trying to cut as literally as we can with no BS to seeing whether they'll really be good:
1. Ask for a 15-minute video coding or going through code they've written
2. Show them a 1-hour video of me going through our own codebase, see what they say
3. Send a repo which is a "slice" of our codebase with an introduction and some real tasks in a readme
I've found:
- Adding them to this demo repo, answering questions and reviewing work for each candidate takes 2-3 hours overall (as much time as organising a video interview) but it's actually 100% them doing what they'd be doing with us
- Main consideration is that we have to pay for these trial hours
- They get to meet me and I get to meet them through our demo videos and even though they're not "realtime" I think we get more chill and authentic impressions of each other than in an awkward call which kinda tests how you do under pressure more than anything
- A significant amount of candidates who would pass the degree/cover letter/past projects etc. test fail at the first step. They may say anything they've worked on is confidential - which is fair enough - but then when we ask for any evidence of a personal project, open source contribution, personal experimentation, even just something they might whip up for us in 15-minutes - they give us a blank look [via email] and yeah maybe it's harsh but at that point it feels like they've been blagging it or working for a big company where they didn't really build stuff themselves
- So far we've trialled 3 developers with the "codebase slice" who looked great on paper, 2 cases delivered mostly AI slop and 1 case was clean code but no attention to detail on the front-end and the result looked patchy and incomplete
(and if you are interested email me at [email protected] - fullstack remote role, fintech SaaS, lots of juicy complex logic and custom components, next.js/Express/Mongoose/Redux)
0 comments
—
DIY hiring methods for a startup in a world complicated by AI · HackerTrans
The traditional methods always feel like a guessing game, whether any of these will actually translate to being a good coder (and a nice human, etc. but let's say this is already part of being a "good coder")
- Uni degrees - Resumes and cover letters - Video interviews - Test coding questions - Past projects
Even before AI - none of these are actually demonstrating that you're a good coder because what you're doing here isn't literally demonstrating that you're a good coder, it's showing something else that's trying to suggest that you're a good coder, right? So there's like, a bridge to jump between saying "Ok, you've written an impressive letter or gotten this impressive degree or pointed to this past project" and from there we need to guess whether you'll actually be good.
And with AI in the picture - things like cover letters, etc. are almost meaningless. Often even past projects as well, we see a lot of "impressive projects they built themselves" which are actually facades - impressive landing pages likely built by AI and then everything falls apart once you actually try and use the web app, "oh yeah registration isn't working currently"
Our approach so far has been trying to cut as literally as we can with no BS to seeing whether they'll really be good:
1. Ask for a 15-minute video coding or going through code they've written 2. Show them a 1-hour video of me going through our own codebase, see what they say 3. Send a repo which is a "slice" of our codebase with an introduction and some real tasks in a readme
I've found:
- Adding them to this demo repo, answering questions and reviewing work for each candidate takes 2-3 hours overall (as much time as organising a video interview) but it's actually 100% them doing what they'd be doing with us
- Main consideration is that we have to pay for these trial hours
- They get to meet me and I get to meet them through our demo videos and even though they're not "realtime" I think we get more chill and authentic impressions of each other than in an awkward call which kinda tests how you do under pressure more than anything
- A significant amount of candidates who would pass the degree/cover letter/past projects etc. test fail at the first step. They may say anything they've worked on is confidential - which is fair enough - but then when we ask for any evidence of a personal project, open source contribution, personal experimentation, even just something they might whip up for us in 15-minutes - they give us a blank look [via email] and yeah maybe it's harsh but at that point it feels like they've been blagging it or working for a big company where they didn't really build stuff themselves
- So far we've trialled 3 developers with the "codebase slice" who looked great on paper, 2 cases delivered mostly AI slop and 1 case was clean code but no attention to detail on the front-end and the result looked patchy and incomplete
(and if you are interested email me at [email protected] - fullstack remote role, fintech SaaS, lots of juicy complex logic and custom components, next.js/Express/Mongoose/Redux)