Along with (as was the main point of the original blog post) probably quite a few false negatives, Like the author of that post, for example.
Meanwhile, you can achieve the same filtering benefit by far less obnoxious and more accurate means (a cursory glance at their repo, fore example).
When your organization can recognize a poor investment but cannot act on it, you have an organizational problem. Why would someone get to hire more "bad" (for whatever that means) if you can recognize it? Even if you're doing all the hiring yourself, it's unlikely you'll have a 100% record over time.
For another, even if "an organisational problem" or inconvenient labour laws don't get in your way, firing employees can have a serious effect on the morale of those left behind. And for good reason! Usually companies that regularly fire people for poor performance has a different sort of culture than companies that don't. In my experience, that's the worse organisational problem.
The right solution is just to be more careful about who you hire in the first place.
Right. But we can't talk about that, of course.
Instead the corrective strategy seems to be: "Let's torture some more candidates. It's cheap and it's easy, after all."
Everyone likes to think they are the good candidate and not the bad one that slipped through…
We have flat hierarchy. Just volunteer to interview because your voice is on the same level as others.
So if someone lied their way through interviews, in the first couple of months you should be able to detect that and terminate their trial period.
Managers hates firing people, it is really uncomfortable. So instead of firing this person to look for a new, they keep the person and say they need even more headcount and hope the next one is better.
It doesn't have to be a public repo of course. They can just send zip file.
You need an equal comparison across all candidates.
"We'd like to see a code sample -- something we can build and compile, that you feel is representative of your skill level."
Sounds pretty straightforward and egalitarian to me.
Long before open source was a thing, and repos were even thought of -- I always had a few 1,000+ line personal projects I could dig up and show people.
1,000 lines isn't that large, btw - but yeah, maybe 500 would be a better number.
With Leetcode, everyone starts on equal footing (there's no bias towards people who code in their free time)
Yeah, it's extremely irritating as a candidate but it's something
>usually being close to runnable with a few small bugs was a good submission and runnable with no bugs was a great submission
This is exactly what we do: check the solution, and if it's close we continue the hiring process. There is a huge difference between an almost working solution and what you get 70% of the time: code that will never, ever, ever work, even if it could run.
A lot of hiring managers have no concept of experiment design.
You can’t invent a measure (or copy it from a blogs about how other companies hire), look at who it rejects, and conclude they are bad. You don’t know if they were good or bad. You never worked with any of them. You only know that you applied your measure.
The only candidates you have any information about are the ones you hired, yet you are insulting the others on the internet today
What's interesting is that the people doing all this "weeding out" seem so sure that they're bravely defending the fort against an onslaught of incompetent hordes... and yet they're missing this extremely basic point about the nature of the process they've chosen to use.
How so?
> turnover rate on your team like?
2 years since I joined, hired one dev recently, no one left so far.
I'd say much has changed since he wrote that, and even for the time it contains some odd advice which he tries to force down the throat of the reader, but I'd be more interested to hear other's thoughts on the article.
He also had this, what seems to me, nightmarish, dev. speed and progress estimation tracking app he made. I don't remember what blog post, but he described it and jokingly wrote that Lisa needed to get the speed up, or something. Fast forward to today and it is a predecessor to our agile nightmare. But I don't think that was his intent.
https://www.joelonsoftware.com/2007/10/26/evidence-based-sch...
The general problem with Scrum like velocity tracking is that devs will start to lie to hit the estimates to keep the commissars of their backs. The best liar gets the straightest graph. And as devs learns to lie, you get the impression of improved estimation.
"Some developers (like Milton in this picture) may be causing problems because their ship dates are so uncertain: they need to work on learning to estimate better."
I mean ... why God. I don't know but maybe poor Milton has the hardest tasks, does not cut corners or what not. Brandon seems to "get it" and has silly narrow error range - I wonder how ...
(I am not bitter. Pinky swear)
Perhaps it's just me, but I took a look and I'm not convinced. Nothing jumps out to me as overwhelming proof of programming ability.
I see that on paper they're a great candidate, but the reason that companies do these tests is that often the candidate doesn't measure up to the resume
Are you suggesting that leetcode tests provide "overwhelming proof of programming ability"?
Leetcode and coding ability is an A -> B but B -/-> A situation: failing leetcode doesn't mean you can't program, but the reverse is absolutely the case.
Leetcode can't filter out other potential issues that may impact success (e.g. poor anger management, or poor judgement).
I should perhaps be clear that by "overwhelming proof of ability" I meant "the ability to program well". Which by definition does intersect with one's capacity for judgement, on a certain level. (Though is largely orthogonal to other behavioral issues, I find).
I see LC as providing proof of, at best, minimal ability. And I've seen plenty of people who can run circles around LC-style tasks, but when it gets to the bigger picture -- you know, writing good, maintainable software -- somehow, it just never comes together right.
We agree that obtaining proof of minimal ability is important. I just think that proof can be obtained in other, far less mutually degrading (and error-prone) ways.
When looking at this person's Github page, I -unlike you- see a vast number of trivial projects that can best be described as regurgitations of other people's work. Even worse, if you actually look at his first pinned project (https://github.com/kevinburke/nacl), you'll see that it is nothing but a wrapper with trivial changes/updates... an anti-project.
In the end, I see code that provides the illusion of competence but, when actually inspected, is actually telling me not to hire this person. The fact that pretty much anyone can declare oneself "a software engineer" today and easily present a veneer of competence together with powerful financial incentives, makes it absolutely necessary for any serious engineering team to have objective measures in place in order to evaluate candidates.
There are so many terrible engineers out there (and some even slip through the cracks and land at places like FAANG) that doing anything less is irresponsible.
But rather, if you're looking for a basic smoke test which can answer the question "Can this person actually code? Do I want to spend more time engaging them?" -- there are better ways to do that than expecting them to cram on a list of algorithms to recite over the phone. Like looking at an actual work sample.
The fact that you looked at this work sample -- and came to your own conclusion about whether they would be worth engaging -- validates this very simple and basic point.
Which is your call of course. He may not meet your bar, but it would be absurd to say this guy can't code, and hasn't been around the block in terms of general CS concepts.
That said -- my own subjective opinion is that you're being exquisitely savage in your assessment of his NaCl implementation. Yeah, it's wrapper -- like it says in the Readme, and like a lot of working code out there. It's a pretty established way of solving a problem, in fact -- in some cases a very efficient and elegant way. It's not the same thing as writing 20k of primitives from scratch, of course. But that's not what he's presenting it as. He's not saying it's a "tour de force". He's saying "I think this shows we can talk about other things besides FizzBuzz".
And then other things you're like saying... like "an anti-project"? Wow.
This is also what a lot of those engineering interviews are trying to ascertain, in a quantitative manner. I couldn't care less about someone's degrees or ivy league schools or Github projects (in isolation). These are all secondary -if that- considerations, problem solving ability being primary.
Using the example you disparaged (his NaCl implementation), I would say:
"On first appearances, clearly yes. The problem was to adapt the a set of standard Go functions to a foreign interface - something one does quite a lot in an engineering environment. Also, the fact that he's aware of projects like NaCl and seeks to learn from the likes of Bernstein and Lange is a positive signal."
If I wanted to, perhaps I could drill down into the code and/or the NaCl spec itself. I don't know Go, so for all I know, maybe his code is actually horseshit. But all I'm saying is - from first appearances, definitely a positive signal. Infinitely more informative that FizzBuzzing (which would insult both his intelligence and mine).
This is also what a lot of those engineering interviews are trying to ascertain.
I've already made it clear that a glance at a repo does not obviate the need for an actual interview.
I guess in your eyes I should have also done the cryptography, however, I am not a cryptographer or a mathematician and one of the first things that you learn when you start doing work with cryptography is that you shouldn't roll your own crypto.
I decided to make that library prominent because in my consulting practice I saw a lot of clients using cryptographic primitives that were clunky and/or had a lot more overhead, for things like cookies or public key authentication or just sharing messages. nacl is difficult to use incorrectly, and is fast and efficient in terms of overhead. I wanted to promote use of nacl more than the work that I was doing for it in particular.
For that client (a smart lock company), I rewrote the backend, that someone else wrote in C for a specification that did not match the needs of the business. The C software was extremely difficult to read and maintain - socket writes in a callback in one part of the code and socket reads in a completely different part of the code, shelling out to curl, simple Mongo queries that would take 50 lines to set up. As a result the engineering team could hardly make any progress deploying new features. I ended up rewriting the whole backend in Go, a few thousand lines of understanding what the C code was doing, and porting it, to the point where the team could actually release patches and new features on a regular cadence, and figure out what on earth was going on with the Mongo queries and shell-to-curl.
That's not public though, if you're looking for public work, you could look at Logrole (github.com/kevinburke/logrole), a HTML client for your Twilio call, SMS, phone number, and recording data. Of course, you could say "that's just a regurgitation of whoever did the API client", which is true-ish, but I also wrote the API client (github.com/kevinburke/twilio-go). Of course, you could say "that's just a regurgitation of the Twilio API," but I also helped build and maintain the Twilio API, as well as the documentation and information architecture (https://www.youtube.com/watch?v=sQP_hUNCrcE), and the client libraries (https://www.youtube.com/watch?v=C_UJHqR_2Mo).
Most recently I was hired as the first engineer at Meter (meter.com) where I built out the entire backend API, the deployment infrastructure, the CI/CD infrastructure, a pipeline for the in-the-field servers to update themselves, and a method for access points to retrieve configuration from the backend. All in all this was probably about 30-50,000 lines of code over two years. I think the API has had one customer facing error in two years and no downtime.
It is true that anyone can call themselves "a software engineer" but not everyone can get contracts for a sole member consultancy and, crucially, get their contract renewed multiple times at every single client. If I was selling snake oil and not delivering anything, that would not have happened.
Looks like he has ADHD. Maybe somebody with more time can go through his blog post line-by-line and figure out what's going on.
Most of my Coderpad stuff compiles and runs the first time, so I don't use his incremental test method, or see a need for it. Maybe he learned programming on heavy-weight IDEs and got trapped mentally in those?
That sounds ad-hominem. I found the post perfectly readable.
Maybe he learned programming on heavy-weight IDEs and got trapped mentally in those?
Maybe he just thinks differently than you do? His body of work is not suggestive of someone "trapped" in an IDE or other dysfunctional midset.
I’m very capable of focusing when there is a lot of money on the line.
As someone who had courses in C on paper, i can only say that it's a colossal waste of time. Tooling exists to make lives easier, gain time and productivity, why avoid it?
It'd be like saying only the farmer who can do everything by hand without any tools increasing their productivity is a real one. Or an accountant refusing to use Excel/calcuators/ERPs and writing everything on paper.
And before anyone jumps to conclusions, i often write small scripts with vim with a minimal setup, but when I'm doing anything complex i prefer to use an IDE
You're trying to map manual work onto engineering.
And absolutely I'd rather work with accountnant who can write me on paper without Excel/calculator/ERP his logic.
It's like you're being obtuse just for the sake of the argument. Someone who can clearly express themselves without specific medium, be it paper, IDE or whatever is strictly superior to someone who is tied to one specific medium.
How does it actively helps and avoids small human errors in reasoning.
> I'd prefer the person knowing using it instead of sticking an old and imperfect method out of pride.
What does even pride have to do with this?
Because reasoning is only a part of everything to be done. An IDE can help with typos, best practices ( global variables, lack of error checking, etc) which are tedious things that take time. Isn't it better to the things that matter instead?
Which applies to everyone? Take IDE out of equation and what do you have?
I'm just like them, only my ide is notepad++.
My code never compiles on first attempt, and never runs correctly on the first attempt.
Just looking at this guy's repo and his writing style -- you'd be foolish to waste his time or yours on leetcode-style hazing. Your conversation should be strictly high-level. If you must, pick a file at random from one his larger projects, and ask a few straightforward quetions ("I'm new to Go - can you explain why I would use this datastructure here?") just to make sure the repo isn't a copy of someone else's, and he isn't outright trolling you. Apologizing as you're doing so, of course ("Sorry, but do you mind if I ask a silly-sounding question or two about something in your repo?").
That's all you need to do. Most candidates won't have work samples that are quite as slam-dunk, of course. But still, your own "smell test" can save you tons and tons of time (and the candidate, agony). It's really, really hard to fake one's way through questions like these -- and really easy to tell if someone is trying to do so.
Put another way, your basic mindset should be: "Trust, but verify". Not: "Joel Spolsky wrote that post 20 years ago claiming that 99.5 percent of all candidates are utterly incompetent. Therefore I need to torture this guy until he confesses."
At least as far as answering the "can he/she code?" question is concerned.