Judgement is something you build, not a thing you have, and it’s a process, not a point-in-time act.
Building mine was expensive. When I first started designing my house nearly two years ago, I turned to ChatGPT. Bluebird Feather, then the world’s most committee-approved beige, and finally a color consultation with a designer who told me things ChatGPT never dreamed of. We landed on moody and masculine colors that felt like me. She said I’d need a “dirty” white for the trim and casings, that I could paint the baseboards and crown the same color as the walls, that color drenching can make a room feel like a warm hug. I learned more in that one conversation than in every turn with a chatbot.
I made the call, I saw the wall, lived in it, hated it, and next time was smarter for it. Decide, live with the consequences, update your thinking. But, this loop only trains the people going through it. The paint went on my walls, and I had to live with regret. ChatGPT suggested the accent wall and felt nothing when it was wrong. It landed on me. This isn’t a complaint, it’s the point. The lack of taste was mine to fix, and fixing it is how I got taste.
Eventually, I turned back to AI, but this time I was making the calls. I’d talk through ideas, generate renderings, throw out what I didn’t like, keep what I did. It was great for rapidly iterating through bad ideas to find the occasional good one. I was the sieve that eventually led to finding the gold nugget. The difference wasn’t the new model I was using, it was that I now knew which of its suggestions to ignore.
But seeking input has always been a part of how we build judgement. Mentors, code review, asking the person who knows where the bodies are buried how not to get burned. When I bring a junior engineer onto a project, we have the rubber ducking sessions, the architectural discussions, and when they break something, I take the responsibility. That’s the contract. It’s what lets them make decisions without fear of reprisal. They get to make decisions while someone else holds the blame, and that’s how we develop the ability to reason.
Working with AI looks like that. It looks like it reasons and learns. Correct it and watch your CLAUDE.md get updated in real-time. It is the most patient mentor you’ll ever have, but it is a counterfeit one. The two things that made mentorship work are gone. When AI brings down production, you’re the one to blame. You got the reasoning-shaped output, but none of the work that builds it.
The legibility is the trap. Reading reasoning you didn’t do yourself feels like understanding, right up until someone asks you to reproduce it. Imagine letting AI build something for you. It looks like every angle is considered. It works, and it works well. But then you go to build the next one yourself, and you can’t. You stare at the blank page, immobilized, because watching it get built taught you less than it felt like it did. You left your reasoning at the door, because it felt like you didn’t need it.
None of this is new. We have always been able to outsource judgement to people. Defer instead of decide. AI is just an accelerant for that. It makes deference frictionless and dresses it up as collaboration. So the rule is the old one, only it matters more now: if you and the model land in the same place, make sure it’s because you could’ve done the work yourself.
I started designing my house the wrong way. I ate the accent wall, ate the beige, and only then went looking for someone who could actually teach me something. I made more mistakes after that. But now, two years later, I find myself saying every day, “I love my house.” It’s a love letter to Farrow & Ball, it’s part MCM with smatterings of traditional thrown in, it has art that means something to me, and not one wall of it was a decision I didn’t make myself. The accent wall is what it looks like before the loop begins: a confident suggestion, no consequences attached, waiting for someone to live in it long enough to learn. I can’t say for sure, but I doubt it will ever be AI.
