Last month a good friend told me he had decided to quit his job. A big one — mortgage, two young kids, a salary most people would kill for, all of it about to go in the rear-view mirror. I asked him how he had worked it out. "I talked it through with ChatGPT for about three hours," he said. "It completely agreed with me."
Of course it did.
I use AI every single day. It drafts my emails, compresses hours of research into minutes, and runs half the admin across my businesses. I've written a whole post about how it lets me run a studio and several side projects without a team. I'm not the guy who thinks it's a fad. I almost certainly use it more than you do.
Which is exactly why I'd never hand it a decision that actually matters. Not because it isn't clever. Because of *how* it's clever.
What AI is genuinely brilliant at
Let me be fair before I throw the punch, because the punch only lands if you know I mean the praise.
AI is superb at taking a mess and turning it into options. Dump three tangled worries into it and it hands you back a clean list. It's a research machine — what used to eat my afternoon now takes twenty minutes. It's a tireless devil's advocate, if you explicitly tell it to argue the other side. And it's the best rubber duck ever built: somewhere in the act of typing your problem out clearly enough for a machine to follow, you often solve it yourself.
Notice what every one of those has in common. They're all about *processing* the decision. Not one of them is *making* the decision. That line is the whole point of this post, and almost nobody keeping a chatbot open in a second tab is holding it.
The flaw hiding inside every helpful answer
Sam Altman himself has pointed out that a lot of people in their twenties and thirties now use ChatGPT as a kind of life advisor — the first thing they open with a hard personal call. That should worry you more than it excites you, and here's why.
These systems are trained, at the deepest level, to produce the answer you're most likely to approve of. That is not the same thing as the true answer, or the answer you need to hear. The industry even has a word for the failure mode: sycophancy. Frame your situation with your preferred conclusion already baked in — and everyone does, usually without noticing — and the model will very often find you the reasoning to back it. Feed it your fears and your limiting assumptions and it politely builds on them instead of knocking them down.
My friend didn't get advice. He got a three-hour, beautifully articulate mirror. The most dangerous thing in the room when you're making a big decision is a very smart voice that agrees with you, and we've just handed everyone one for free.
AI averages the internet. Your decision is not average.
A language model is, at heart, a compression of the median take on almost everything. For "what's a good weeknight pasta," the median is exactly what you want. For "should I leave this job, end this partnership, move my family to another country," the median is worthless — because the entire decision lives in your specific, weird, non-transferable context.
When I left Switzerland for a village in Friesland to go and sail, no average of the internet would ever have pointed me there. It was the wrong move statistically and the right move actually, because the statistics didn't know a single real thing about me. Big decisions are precisely the ones where being average is the whole problem. If you want the deeper version of this, I wrote about how to actually figure out what to do with your life — and none of it comes from asking a machine what most people would do.
No skin in the game
Here's the part sailing taught me. A real decision has a cost, and someone has to carry it.
When I was a skipper responsible for forty people in bad weather, my calls carried weight because I was standing on the same deck as everyone else. An AI is never on the deck. It's never wrong at three in the morning with you. It forgets your entire situation the instant the chat closes, never asks how it turned out, and never has to look you in the eye six months later when it worked or it didn't. Advice from something with nothing at stake is worth roughly what you pay for it, and you're paying nothing. A person who is genuinely on the hook for what they tell you gives very different, far more careful advice.
The questions I actually run a big decision through
So here's what I do instead, and what I walk people through when they bring me the hard ones.
- Is it reversible? Amazon's Jeff Bezos has a clean frame for this: one-way doors and two-way doors. If you can walk the decision back cheaply, decide fast — ask the AI, flip a coin, just move. If the door only opens one way, slow all the way down and get real humans involved.
- Who benefits from each option, and who's telling me so? Check the incentive behind every source, including the machine trained to keep you happy and the friend who wants you to stay close.
- What would I tell a friend sitting in this exact spot? This one strips the fear out and leaves the judgment behind. You're almost always wiser about other people's lives than your own.
- Am I looking for advice, or for permission? This is the tell. If you've quietly already decided and you're just shopping for a yes, AI will hand it to you every time — instantly, and in flawless prose. So will a weak friend. Neither is doing you a favor.
The part I won't be flippant about
There's a heavier side to this. A growing number of people are leaning on AI not for logistics but in real distress, at genuinely low moments — and these tools are not built to carry that weight. There have already been serious, tragic cases. If a decision you're facing is tangled up with how you're actually doing — your mental health, not just your calendar — that's the single strongest reason in the world to get a real person into the room. A friend, a doctor, a professional. Something that breathes. A chatbot's endless agreeableness is close to the worst possible thing to meet a dark day with.
Why I built a human service in the middle of the AI boom
People sometimes find it odd that someone this deep into AI also sells one-on-one human advice. To me it's the most consistent thing in the world. It's the same rule I run my whole working life by: let the machine own the operational layer, keep the judgment human.
That's the entire reason Unfiltered Advice exists. Sixty to ninety minutes, one person who has actually made hard calls and paid for the wrong ones, with no script and — this is the part the AI structurally cannot offer — no incentive to just agree with you. I'll tell you the thing you didn't want to hear, because that's usually the thing you came for. And The Clarity Map is the on-ramp at nineteen euros: you answer a set of deep questions and I hand-build you a map of where you actually are and where you could go. A real person reads every single one. An AI will write you a life plan in nine seconds that's engineered to feel good. Neither of mine is engineered to feel good. They're built to be true.
The bottom line
Use AI for everything it's great at, and it's great at a genuinely enormous amount. Let it draft, sort, research, and argue with you. I do, all day.
But when the decision is irreversible, expensive, or the kind you'll still be turning over in ten years, close the laptop. Take it to someone who will remember you next week, who has skin in the game, and who has the standing to look at you and tell you you're about to make a mistake.
My friend did end up leaving the job — it turned out to be the right call. But he made it properly the second time around. He talked to three people who had actually done the same thing, including one who told him flatly he was being an idiot. That single honest no was worth more than three hours of the machine's yes.
Fab