Let’s talk about the one thing you know about your own street that no machine will ever know.
There is this moment, and if you have lived anywhere long enough you will recognise it. Rain’s been coming down for an hour. Someone’s about to drive out and asks you, quick, will the road be okay. And you answer without checking a thing. Not that way. Take the top road. The bottom one holds water at the dip just past the third gate — every single time.
The map on the phone says the road is fine. You know it isn’t, because you were eleven years old in that dip, and you got sent home wet in your school shoes, and nobody ever had to explain it to you after that.
Nobody taught you that. You absorbed it the way you absorbed your mother’s kitchen. By being in it. By getting it wrong once.
Now hold it next to what everyone is building.
An AI model carries the general shape of every place on earth at once. That is a real trick, and a big one. It can tell you, statistically, which parts of your city flood. It can rank neighbourhoods, price property, route couriers, predict traffic. It knows everywhere.
What it cannot do is know your dip.
Because the model isn’t built from places. It’s built from the average of places. Every road is an ideal road, every field is a median field, every customer is a segment, and the particular gets smoothed out until it reads as noise. Noise is exactly what the system is trained to ignore.
And here’s the sentence people need to hear clearly, because everything downstream of it is a lie:
Local knowledge is just data we haven’t collected yet.
That’s the pitch. It’s what every platform means, quietly, when it tells you the future is one big shared intelligence that will finally get your town right. And it’s wrong — not because the collection is hard, but because local knowledge was never a collection problem. It’s a relationship.
Think about the older shopkeeper. Or the oke who has farmed the same piece of ground for thirty years. What they hold isn’t a dataset. It’s an account of a place they’re answerable to. The farmer knows the northern block drains two days slower and you plant it last, every season, no matter what the calendar says. The shopkeeper knows which of her regulars can be carried until Friday and which one never comes back if you say no.
None of that lives in a file. It lives in a person who has to live with the consequences of being wrong about it. That’s the part the machine doesn’t have. Not the information — the consequence. It can know every fact about your town and owe nothing to it. It has no skin in the outcome, so “getting it wrong” costs it nothing.
And what we do, as an economy, is take the person who knows the dip and replace them with the system that knows the average — and call it efficiency.
It is efficiency. It’s also the slow deletion of the only thing that made the place work.
Watch what leaves when it goes. The local supplier who knew your account gets replaced by a procurement catalogue. Cheaper, faster, and now you’re a line item. The doctor who knew your family’s history gets replaced by a triage model. Quicker, and you’re a risk score. The neighbour who’d have noticed your gate standing open gets replaced by a camera feed nobody is actually watching.
Every one of those swaps is sold as an upgrade, and in a narrow sense each one is. What we don’t price is the thing that walks out of the room when the person does. Not their labour.
Their attachment.
Because attachment was never a cost of doing business. It was the business. A person who is answerable to a place makes different choices than a system optimising a place.
Here’s the human consequence, and it isn’t abstract.
When the machine knows everywhere, it starts treating places as interchangeable. Land becomes a site. A town becomes a labour pool. A farm becomes a line on a spreadsheet — and the buyer can pay ten times the value so it’s hard to say no, and nobody in that whole transaction ever has to stand on the soil afterwards.
A person who is from somewhere can’t do that. They have to keep living next to what they did. That is the whole test. Not “is the model accurate.” Of course it’s accurate. Ask the harder question: who is left standing in the room who has to live with the outcome?
That is what a place actually is. Not coordinates. A group of people who have to keep living next to what they did.
So here’s the operator lesson, because this can’t just be nostalgia.
Whatever you build or sell, the part of it a competitor can’t copy with an API call is the specific, accumulated, answerable knowledge of your actual customers in your actual place. Not your “best practice” — that’s boilerplate the entire market shares. Your particular version of it. The thing you only know because you were there long enough to be wrong and then had to fix it.
Write that down before you lose it. Hand it to the next person who’s going to be answerable for it. When a tool offers to flatten it into a generic workflow, take the tool — but keep the dip.
Because the moment your business can be run by anyone with the same software, you’ve stopped being from anywhere. You’ve become the average of yourself. And an average of yourself is a very easy thing to replace.
Machines are going to keep getting better at knowing everywhere. That isn’t the fight. The fight is whether the particular survives the general — whether the person who knows their street, their field, their customers, their dip, still counts for something in a world that can hold all of it at once and care about none of it.
The principle that must survive is easy to say and hard to hold:
The right to be from somewhere. To belong to a place, to be needed by it, and to have that belonging still mean something when a model that has never stood in the water claims to know your road better than you do.
You’ve got this.