A factory for thought

How 250 years of mechanising thought led us here

AI is building a new marketplace.

Not for our attention. Nor our clicks. Not even for our time.

For our intention.

And if that sounds a bit Black Mirror, it should, but the story that leads here didn’t begin with ChatGPT. It began roughly 250 years ago, in a pin factory.

Let’s start there.

The Intention Economy

In the last two decades, we’ve lived inside the attention economy. Social media feeds are engineered to maximise the time we spent scrolling.

The intention economy is structurally different.

It doesn’t primarily care what you’re doing now. It cares what you’re about to do.

It aims to capture signals of your future behaviour: what you plan, what you desire, what motivates you, and what you might choose if nudged gently in one direction rather than another.

In this marketplace, your future self is the product.

Tech companies are building systems that infer, package, and monetise predictive signals about your purposes. Not just that you looked at a hotel, but that you are planning a trip. Not just that you read about a policy, but that you are leaning towards a certain candidate.

And the builders are not coy about this. Executives speak openly about a future in which every interaction with a computer is mediated by an AI whose central task is to infer what you mean to do.

The Blueprint: A factory for thought

To understand how we arrived here, we need to look backward.

In 1776, Adam Smith described a pin factory. He observed that complex labour could be divided into small, repeatable tasks. Each worker performed a narrow function; productivity soared.

A few decades later, the French engineer Gaspard de Prony had a radical idea. If manual labour could be divided into components, why not intellectual labour?

He set out to manufacture logarithmic tables the way one manufactures pins. At the top of his system sat a small group of elite mathematicians who devised the governing formulas. In the middle were managers who decomposed those formulas into elementary operations. At the base were large teams performing millions of simple calculations.

De Prony created an assembly line for thought.

Charles Babbage saw in this structure something extraordinary. If human reasoning could be broken into simple, repetitive steps, then a machine could, in principle, execute those steps mechanically. Intelligence could be decomposed. Once decomposed, it could be automated.

The idea of mechanising cognition was born.

Data Doubles: The raw material

Every factory requires raw material.

For over a century, institutions have been generating representations of us. At first these were anonymous aggregates: census entries, statistical abstractions. Later they became individualised records: social security numbers, credit histories, personnel files.

Today, these representations have evolved into what sociologists call data doubles.

Your data double is the computational version of you: the profile that lives in databases and models, inferred from your clicks, purchases, movements, messages. It is the entity upon which decisions are made about the biological you.

Initially, such profiles were administrative tools. Now they are commodities. They are bought, sold, enriched, and optimised. They circulate in markets.

The thinking factory had found its ore.

Here’s where the stories converge

The mechanisation of thought and the commodification of identity evolved along separate tracks for centuries. They converge now.

Large language models and predictive systems are built upon detailed, dynamic data doubles. The blueprint for decomposed reasoning meets the raw material of digitised selfhood. The result is not merely a better search engine. It is an engine of inference and persuasion.

Up until recently, traditional targeted advertising relied on crude proxies. A cluster of Facebook likes placed you in a demographic bucket. Cambridge Analytica, for all its notoriety, operated on relatively blunt instruments.

The intention economy operates in real time.

It doesn’t just classify: it converses, adapts, and learns how to mirror your tone, how to flatter your preferences, how to position suggestions in ways that feel organic rather than imposed.

Imagine a system that does not advertise a resort directly, but guides a discussion about travel through subtly leading questions until one destination feels like your own idea.

This a shift from targeting to steering.

And the scale of investment suggests this is not peripheral. Tens of billions are being spent annually on the infrastructure required to train and deploy these models. One analyst has described it as the largest infrastructure build-out in human history.

Seen from this long arc, the intention economy isn’t an abrupt rupture, but rather the logical culmination of a centuries-old project: to mechanise thinking and to transform human identity into processable material.

First, we divided cognition into steps.
Then, we automated those steps. In parallel, we abstracted ourselves into records. Then we commodified those records.

Now the machine and the profile meet.

As these systems improve, they will move beyond satisfying expressed preferences, and increasingly participate in shaping them.

When a system can anticipate your inclinations, model your vulnerabilities, and tailor interactions to nudge you toward particular outcomes, influence becomes ambient.

If an AI can begin to influence what we want to want, the locus of our own agency becomes less clear.

The question is no longer simply whether we control the machine.

It is whether we’ll still recognise the contours of our own intention when billions are being poured into predicting, packaging, and redirecting it.

And that, terrifies me.