Table of Contents
The brush paints itself
Last week I watched The AI Doc: Or How I Became an Apocaloptimist. Daniel Roher set out to make a film about AI, found out early on that he was going to be a father, and the film became his search for what kind of world his son would grow up in. As a father myself, I watched it for much the same reason: I wanted someone to tell me it would be fine.
Nobody did. What surprised me first was how many of the people building this technology agreed to be interviewed. What surprised me more was how few of them claimed to know where it’s going. Ask about the next three to five years and the answer is some version of “we’re still figuring it out”, or so they say. Ask whether anyone is prepared to slow down and the answer is no. Safety testing, regulation and international cooperation are all running well behind the technology itself.
Our team has worked on AI projects since before ChatGPT, back when we called it machine learning. Even so, the volume of releases, claims and noise over the last two years is hard to keep up with. That’s partly why we started this newsletter. It’s also why we’ve noticed some recent pieces have become too technical. So this is the first of a different kind of post. It’s less about how the tools work and more about where we think this is heading, in plain English.
Here’s where I’ve landed for now.
Give Velázquez and a child the same brushes, oils and canvas and you get two very different paintings. The tools are the same. The judgement isn’t. That has been true of every technology so far.
What’s new is that the brush has started painting by itself. Hand the child a brush that produces a competent canvas on its own and the result will look fine. Most people won’t notice anything. Velázquez would see what’s wrong in seconds.
That’s the shift. AI lowers the expertise needed to produce an answer. It doesn’t lower the expertise needed to recognise a good one. As producing gets cheaper, recognising becomes the scarce skill.
For anyone running a space, this is already practical. AI can draft your membership terms, rewrite your pricing page, reply to a member complaint or summarise last quarter’s occupancy. All of it will read well. But for anything that matters, someone who knows the business still has to read it before it goes out. Do the terms protect you? Does the pricing suit your market? Does the reply calm that member down or wind them up? The risk isn’t output that’s obviously bad. It’s output that’s plausible and that nobody checks.
Final thoughts
Nobody knows what AI looks like in three to five years, including the people building it. Plan for uncertainty, not for anyone’s prediction.
AI makes producing easy. It doesn’t make judging easy. That’s where your value and your team’s value now sit.
The new risk is work that sounds right but is wrong. Match the checking to the stakes. Routine replies can run on their own, with clear rules for when to hand over to a person. Anything that touches a contract, a price or a member relationship needs someone who knows the business to look first.
Build judgement on purpose. Train people to check AI’s output, not just to prompt it.
I’m not sure any of this makes me an apocaloptimist. But it’s a more useful position than waiting for someone to tell us it will be fine.
