Let’s talk about the difference between being predicted and being hungry.
Those two things get treated as the same thing now, and they are not. One is a record. The other is a motor.
Picture someone at half past ten at night, phone in hand, deciding what to eat. Before they’ve finished the thought, the app has it. Not the dish — the category. It knows they order Thai on Tuesdays when the week has been bad, it knows they go for the cheap option after payday and the expensive one before, it knows they abandon a basket when the delivery estimate passes forty minutes. So it puts the right thing at the top of the screen, and they tap it, and the whole thing takes nine seconds.
That feels like being understood.
It isn’t. It’s being remembered. And there’s a difference between a system that remembers what you did and a person who wants something they haven’t done yet.
Here’s the mechanism, because this is where people go soft and vague.
A recommender models preference. Preference is built from your history — the choices you already made, the ones you can be observed making. It is backward-facing by design. It has to be. There’s nothing else to train on. So what it gets very good at is the shape of your past, projected forward, and the projection is accurate enough that it feels like insight.
Desire is a different object. Desire is the thing you don’t have words for yet. It’s the reason a person walks into a shop knowing they want something and comes out with a thing they couldn’t have described on the way in. It’s the reason a founder builds a product the market research says nobody wants. It’s the reason anybody ever changed their mind about what they wanted to do with their life at forty-one.
You cannot train a model on that, because the training data doesn’t exist until after the person has done it.
Which means the machine is not learning what you want. It’s learning the outer boundary of what you’ve already accepted, and then it’s building a very comfortable room inside that boundary.
Now the part that matters more than the philosophy.
When something is predicted accurately enough, often enough, you stop doing the work of wanting. Not because you’re weak. Because wanting is expensive. It requires you to sit in an unclear state, tolerate not having, and form a preference from incomplete information. That’s effort. And effort gets skipped when there’s a shortcut sitting on your lock screen.
So the muscle that gets used least is the one that decides what’s worth wanting in the first place.
You can watch this happen in companies, and it’s uglier there because it costs money.
Open ten product categories right now. Ten apps, ten brands, ten pitches. They’re converging. Same features, same pricing tiers, same roadmap language, same homepage with the same three-word headline and the same stock photo of a person holding a phone. Everybody’s confused about why nothing feels differentiated anymore, and the answer is sitting in their analytics dashboard.
They all built the same machine. It reads what customers already do, aggregates it, and tells them what to build next. So they build the same next thing. The data is real, the process is rigorous, and the output is the average of everybody else running the same process.
Nobody in that building is allowed to want something. They’re only allowed to forecast.
And forecasting is a fine tool with a fatal blind spot: it can only tell you about demand that already exists. It has no mechanism for the thing that creates new demand, which is a person with an unreasonable appetite and the stubbornness to pursue it before the evidence arrives.
Think about the products that actually moved you. Not the ones that served you well — the ones that changed what you thought you wanted. Almost none of them came out of a demand study. They came out of somebody being bothered by something, or obsessed with something, or refusing to accept that this is just how it is. The data would have said no. The data always says no to the thing that hasn’t happened yet.
So the machine’s blind spot is not intelligence. It’s that it has no stake. It cannot want. It can only estimate what you’ll accept.
That’s the line I’d hold onto, and I’d hold it hard.
A system with no appetite can optimise anything you hand it. It cannot tell you what’s worth doing. It cannot be offended by a bad experience. It cannot fall in love with an idea at two in the morning and spend four years on it. It cannot prefer the harder, better version over the cheaper, adequate one, because preference — the real kind, the kind that costs you something — requires a self that stands to lose.
That’s not a small gap. That’s the whole thing.
And it’s why I think the most valuable thing you can protect in yourself, and in the people who work with you, is not creativity or focus or discipline. It’s appetite. The capacity to want something specific, on your own authority, without a case study.
Practically, that means a few uncomfortable things.
It means keeping at least one thing you want that has no business case attached to it. Not a side project with a landing page. A genuine want — a standard you hold because you decided to, a way of doing the work that costs you margin, a thing you build because the version in the world is insulting and you can do better. If everything you want is justifiable on a spreadsheet, you’ve already been modelled, and you’re just executing the model.
It means occasionally deciding from appetite rather than forecast. Not always — you’d go broke. But the decisions that set the direction of a business are not the decisions analytics should own. Analytics tells you how to get better at what you’re already doing. It cannot tell you whether you should be doing it at all.
And it means protecting the person in the room who wants something weird. The one whose idea doesn’t fit the roadmap, whose instinct you can’t validate, whose proposal makes the spreadsheet look bad. Most companies kill that person politely, in a prioritisation meeting, with a good process. Then they wonder why the only ideas left are the ones the competitors also had.
There’s a version of the future where all of this is very smooth. Every need anticipated, every choice pre-filtered, every product arriving before you knew you wanted it. It’ll be efficient. It might even feel like care.
But a person who never has to want anything is not being served. They’re being managed.
The machine can predict what you’ll take. It has no idea what you’d fight for. That’s still yours, and it’s the only part of you that was ever going to build anything worth keeping.
So keep one thing you want that nobody asked for. Keep it badly, keep it long, and don’t let a dashboard talk you out of it.
You’ve got this.