You're hired

It's no secret that most job interviews are basically seances with better lighting. Some studies indicate that you can predict a lot of the outcome of an interview from "thin slices" as short as minutes or even seconds. Of course, it's impossible that a slice that short could tell you anything important about job ability, so we're left with the conclusion that it's mostly a load of confirmation-bias-driven woo.

In recent years, the tide at technology companies has been turning, though. Google have slowly reformed their famously Byzantine interview process that mostly relied on whiteboard algorithmic pressure tests and clever gotcha puzzles that were mostly for showing off how clever the interviewer is. Apparently their internal data showed their interviews were almost completely unpredictive of job performance, and to their credit they acted to get rid of the bunk and replace it with better-validated methods.

More recently, companies have begun to to hone in on a technique where you interview based on actual work performance. As in, after a certain point in the process you just pick some relatively newbie-friendly task that actually needs to be done and pay the candidate to do it. It's a refinement of the idea of work sample tests, where you use a fake task that's representative of the work they'll be doing. But, hey, why not cut the fakery and just use real work? Apparently it's very predictive.

The whole thing makes me think about a glorious endgame: could you just make a job offer to the whole world? Congratulations, everyone, you're hired. Your job starts whenever you want, just let us know when you're ready and we'll give you a task to do. The rate will be based on how well you've completed previous tasks, which means for the first one it'll be pretty low (but also a very small task). If you do it well the next one will be much higher as the error margin on your pay grade goes down.

Of course, the kind of company structure required to make something like that work would have to be pretty sophisticated. You'd have to have a predictive engine with human oversight that could assign and update expected values on task-person pairs very quickly. You would even need special tooling so that tasks could be broken down into very small chunks, which in the case of software would almost certainly mean some kind of programming language-level support. The hurdles are pretty big, here.

But, still, it seems like it could be a truly fascinating and innovative way to run a business. The entire hiring pipeline and all of the complexity and bias that goes with it only makes sense if there's a meaningful distinction between hired and not hired. Maybe there doesn't need to be.