CAN YOU WRITE SOFTWARE USING ONLINE CODE TESTS?
I was reminded of the unfortunate reality of code tests by way of an
unhappy job applicant on a Discord server. A coder was woe-ing the
hours they had spent on a code test for a job. It got me thinking
about code tests, including my own experiences1 (well… lack
thereof). Then my mind began to speculate on just how sinister code
tests could become. Like, could you get a piece of software written
entirely from code tests? Imagine: a tech company whose product is
software that develops software out of code tests. These code tests
are embedded into job openings for positions in companies that don’t
exist. A code test for one position at one company might involve
writing a front-end template or small API. A code test for another
position at another company might involve testing the results of the
other applicants' work. Coders apply to the jobs and complete the
tests. Regardless of the outcome they’re never hired and get no
further in the process. This cycle repeats in endless iterations. Code
gets written, but no one ever gets paid. All the while–and unbeknownst
to all who’ve applied–a piece of software is being developed behind
the curtain.
Can it be done? Just how difficult would it be to execute this
sinister plot? (And: if I am to reason about how to implement this
scenario, is that in itself a type of code test?) Setting up a fake
company would be the easy part. There are website generators, brand
asset generators, and website copy generators to quickly stitch
together something that looks like yet-another startup company. But
what about the code tests themselves? The scope of this problem is
beyond anything I’ve ever had to reason about. I couldn’t tell you
what language, development kit, platform, etc., to use. It strikes me
that maybe a Lisp dialect would be effective, since I hear they are
well suited for AI, and AI would probably factor into this scenario
somehow. Nonetheless, Clojure (because it’s a Lisp dialect and it can
run on a JVM) and Haskell (for its strongly type nature, and the fact
that its used by Meta for spam detection) would be my language
candidates. (And since I want to learn Haskell, so I guess I’d just go
for that :)) But that approach is all based on the assumption that a
program can be written to intelligently write, evaluate, and stitch
together code tests. And this assumption could very well be flawed.
So instead, I think it would be worth exploring a direction that
leverages version control. I’m become more familiar with Github’s
actions and API, so I think there could be a solution that makes use
of the platform. A piece of software could be written (using any
scripting language, really) to write, pull, merge, lint, and build
code tests. The software could also open new repos which job
applicants get invited into for their test. I’m imagining a piece of
software who’s main job is “perception management”–to keep up the
perception that the code a job applicant is working on hasn’t been
written by a previous job applicant. The software achieves this by
moving the code around between repositories to obfuscate the
author. To kick the whole thing off the first test could be to write a
code test. And the next code test (to be completed by another
applicant for a different “position”) could be to evaluate the
test. The cycle is endless. Broken down into discrete chunks, code
tests could be used, brick by brick, to build a piece of software that
builds software using code tests. Evil stuff! Nobody better get any
ideas…