Here's a fun little existential crisis to ruin your Monday: if AI can write your emails, crunch your data, and spit out code faster than you can finish your coffee, what exactly are you being evaluated on? According to a piece in Fast Company, the answer is basically nothing that matters.
Performance reviews, it turns out, have always been a bit of a theatre production. We dress them up in spreadsheets and quarterly metrics, but what we're really measuring is speed, task completion, and throughput. Which is, coincidentally, exactly what machines are stupidly good at.
The 19% problem nobody wants to talk about
A recent management study cited by Fast Company found that AI can help workers get through 12% more work, 25% faster. On paper, that sounds like a glowing performance review for the robots. Until you hit the footnote: AI gets the answers wrong 19% of the time.
Nineteen percent. That's nearly one in five. Imagine if your accountant was wrong one in five times, or your surgeon, or the GPS that's confidently directing you into a lake. We're not talking about minor typos here - we're talking about a fundamental flaw baked into the very thing that's supposedly making us obsolete.
And yet, we keep optimizing for throughput. More output, faster delivery, higher volume. We've essentially built a race where the finish line is a cliff.
So what are humans actually for?
This is where it gets genuinely interesting, rather than just depressing. The 19% error rate isn't a bug to be fixed eventually - it's a flashing neon sign pointing at the gap that humans are supposed to fill. Judgment. Context. The ability to notice when something is technically an answer but completely, catastrophically wrong.
The problem is that none of that shows up in a performance review. You don't get a line item for "caught the AI's confidently wrong conclusion before it became a press release." There's no KPI for wisdom, for knowing which questions to ask, or for understanding why the numbers feel off even when they look right.
The real crisis isn't AI taking jobs
It's that we've been measuring the wrong things about human work for so long that we've forgotten what the right things even are. AI didn't create this problem - it just arrived with a spotlight and pointed it directly at the awkward mess we've been sitting in for years.
Performance reviews were always a bit performative. Now they're just embarrassingly, undeniably, 19%-error-rate obviously so. Maybe it's time to figure out what we actually want humans to do, before we finish building the system that replaces the version of us we never really wanted to be anyway.
