The redundancy fairy isn't real

Automation, historically, has always been a deskilling machine. Take a job that requires judgment and skill, break it into dumb little steps, then hand each of those steps to whoever's cheapest: a machine, a new hire, someone on the other side of the ocean. Every boss who's ever seen a new tool has run this exact play for at least 2 centuries, because it's the only play that’s ever been in the book.

Generative AI breaks that book. Output quality tracks the skill of whoever is prompting and checking: garbage in - garbage out, at scale.

Give one to someone who can’t already do the job properly and you get fluent, confident nonsense, generated at the speed of light and the scale of the cloud.

Here's what a Large Language Model actually does, in competent, experienced, hands: it can turn 10 skilled workers into the output of 12, occasionally 20. Now and then it turns 10 into 5, because someone's spending more time fixing hallucinations than the tool saves them.

What it cannot do, what it has never once done, anywhere, for anyone, is turn 10 into zero. Someone still has to know enough to catch the model's confident mistakes, and that someone doesn't come out of nowhere.

Every layoff premised on "AI can do that job” is a bet on a technology that hasn't been invented.

Outside a few very narrow lanes, generative AI needs so much adult supervision that today’s CEO fantasies of it replacing workers wholesale is going to look absolutely stupid in retrospect (once you strip out all the layoffs that got blamed on "AI" to cover for the real reasons, of course).

Colin