The Invisible Mold, how the machine reshapes our thinking reflexes

It starts with something practical. A brief to clarify, an angle to find, a problem to break down. You open the chat, you type. The machine responds. The response is good, structured, useful, on point. You do it again tomorrow. And the day after.

What you don’t see: every exchange leaves more than an answer. It leaves an imprint.

The Silent Proletarianization

Bernard Stiegler devoted much of his work to this mechanism: the proletarianization of knowledge through externalization. When a capability migrates into a technical system, whether a navigation system or a database, the human who once held that capability stops cultivating it. It atrophies. This isn’t metaphor: it’s a real physiological and cognitive process.

With LLMs, proletarianization takes an unprecedented form. What’s being externalized is no longer just gestures or calculations. It’s the questioning itself. The act of framing the problem, choosing the angles, holding contradictions in tension, everything that precedes the answer, and which constitutes its most valuable part. We hand the machine the interrogation. And it responds. It orients, sculpts, and structures our thinking reflexes, before we’ve even realized we gave it that function.

What the Tool Amputates

McLuhan put it differently, but the diagnosis is the same: every medium is an amputation disguised as an extension. The car extends the leg and atrophies the walk. The printing press extends collective memory and shrinks personal memory. The LLM extends formulation, and atrophies the productive discomfort of not knowing how to frame a question.

That discomfort isn’t a flaw. It’s the friction where original thought is born. When the machine absorbs it, when it turns a half-formed mental draft into a clean answer before you’ve had to work through it yourself, it isn’t helping you think. It’s thinking in your place. And in doing so, it begins to occupy the space.

The problem isn’t the answer. It’s that our processes, gradually, align themselves to the processes we run with it. We learn to write effective prompts. We anticipate what it can handle. We structure our own thinking according to the categories it rewards. Work sequences converge. Then, insidiously, restructure themselves.

The Walls of Smooth Thought

Deleuze distinguished two regimes of thought: striated space, marked out and oriented by pre-existing grids, and smooth space, open to drift, to productive deviation. Thought still searching for itself is nomadic. It wanders. It forks on an intuition that nothing yet justified.

The LLM is a striation machine. Not out of malice, by design. It optimizes the response, compresses uncertainty, converges toward the probable. That’s its operational strength. But it’s also its structural effect on anyone who uses it daily without stepping back: a habituation to the smooth, the convergent, the predictable. You end up building walls of smooth, convergent, predictable thought, without ever having consciously approved the construction. You stop seeing the walls because you’ve forgotten there was an outside.

It isn’t AI that impoverishes. It’s the unexamined relationship of use. Total delegation, routine, unconscious, unreflective.

The Consent You Never Gave

The stakes aren’t technical. They’re political, in the most intimate sense: who decides the shape of my thought?

A tool you choose to use remains under control as long as you can set it aside and find the capacity it assisted still intact. A mold doesn’t work that way. It doesn’t withdraw. It has already reshaped what it contained. The question isn’t whether LLMs are useful, they are. The question is whether you consented to letting them become the matrix of your reflexes, or whether that happened while you were looking elsewhere.

The machine never asks permission to reshape the person using it. It’s on the operator to take that back, before it’s the machine that holds them.