LLM Outsourcing of Thinking

A simple explanation

There is work the body used to do — sit with a problem, hold its parts, watch the parts rearrange themselves, notice when a thought arrived. That work was slow, often frustrating, and almost always produced a thinker who had grown from the doing. The output was the artefact. The thinker was the deposit.

LLM Outsourcing of Thinking is the progressive routing of that work out of the body and into a model. The artefact still appears. The thinker does not grow. The task is completed and the person who would have grown by completing it has been bypassed. Over time, the bypass becomes the only path the body still knows how to use.

An everyday example

You sit down to write a strategy document. You have, in your head, three rough intuitions about the problem. In a previous version of your life, you would have sat with those intuitions for an hour, found that two of them were wrong, discovered a fourth that none of the three predicted, and finished with a document whose argument you could now defend in conversation because you had built it.

Today, you give the three intuitions to the model. The model produces a structured argument incorporating all of them. You edit lightly and send. The document is competent. The meeting goes well. You notice, when asked a probing question, that you cannot quite reconstruct the reasoning — because the reasoning happened somewhere outside you. The artefact has your name on it. The thinking does not.

Why is my brain feeling slower since I started using AI heavily?

Because thinking is a muscle and muscles atrophy with disuse. The model does not weaken you by doing your thinking; it weakens you by doing it instead of you, repeatedly, in the situations that would have grown the muscle. The Reward System, optimising for task-completion, reads the completed artefact as success and routes the next task through the same channel. The System is not wrong about the artefact. It is wrong about what the task was for.

There is also a more specific cost. The slow form of attention required to hold an unfinished thought is itself an exercised capacity. Each time the model is reached for before the slow form has produced its own answer, the slow form gets less practice. After enough months, the body is no longer comfortable in the state where the answer has not yet arrived — and that state is where most original thinking happens.

The behavioral loop

A loop that operates at the scale of whole cognitive tasks:

1. Task arrival — a problem requiring synthesis, analysis, or judgement.
2. Initial intuition — the body produces a working partial answer or starting orientation.
3. Model recourse — the partial answer is given to the model, often before it has been allowed to consolidate.
4. Mechanical effort — the writer prompts, reads output, refines the prompt, re-prompts. This feels like work.
5. Output assembly — a coherent artefact emerges from the model and is lightly edited.
6. Closure — the task is completed in less time than the unassisted version would have required. The System flags efficiency as success.
7. Cognitive atrophy — the synthesis that would have grown the thinker is never performed. The muscle weakens by one repetition.
8. Default flip — the next task begins with the model, because the body has rehearsed the path so often that the unassisted version no longer feels available.

Emotional drivers

Four feelings, easily confused with productivity:

• A relief from the slowness of unaided thought — the model is unambiguously faster.
• A pride in volume — more tasks completed, more documents produced, more output.
• A faint and chronic foggy feeling that the person cannot quite locate the source of.
• An anxiety, sometimes hidden, about whether you would still be capable of the work if the model were unavailable.

What your nervous system does

The slow cognitive work the body used to do has a parasympathetic signature — a softer, slower attention that allows complex material to consolidate. The model interaction has a different signature — a faster, more sympathetic-tinged engagement of prompting, reading, evaluating. Both feel like work. They are different kinds of work.

Over months of heavy model use, the body's baseline shifts. The capacity for the parasympathetic slow-attention state shrinks. The person becomes someone who finds it physically uncomfortable to sit with an unanswered question for more than a minute. The discomfort is read as boredom or impatience and is relieved by reaching for the model, which deepens the substitution. The shift is small per instance and large across a year.

The DojoWell interpretation

LLM Outsourcing of Thinking is the substitution mechanism operating on the most generative substrate the person has — their own cognition. The Meaning System's original ask was for thinking that produces both an artefact and a thinker. The Reward System's substitute is model-as-cognition — a process that produces the artefact while leaving the thinker unbuilt. They share a surface property: both finish tasks. They are opposite on the inside.

The deposit looks unambiguous on the artefact metric. The document exists. The argument is coherent. The task is closed. The deposit looks closer to zero when measured against the thinker — the synthesis was not performed by the person, the reasoning was not exercised, the practice that would have grown the muscle was outsourced. Over a year, the artefact-count is impressive and the thinker has grown smaller.

This is why the density signature is effort without deposit rather than residue accumulation. The effort is real — prompts are written, outputs are read, refinements are made. The deposit, into the person, is near-zero. The mechanical effort is felt and counted; the cognitive growth is missed. Density is low because the equation is being run on the wrong subject — the model's process, not the person's. The person's cognition received almost nothing for the energy spent.

How do I use AI without losing my ability to think?

You preserve the cognitive tasks that grew you. The category is small and specific. Not every email needs to be written by hand; not every research note needs to be drafted by you. But the tasks that build the thinker — the synthesis problems, the difficult writing, the original questions in your domain — these need to be performed by the person they are supposed to grow.

The work is to identify which tasks were always for the thinker and which were always for the artefact. Outsource the artefact tasks freely. Protect the thinker tasks fiercely. The protection looks like time, quiet, and a willingness to sit with the unanswered question long enough for an answer to surface.

Practical steps

1. Distinguish thinker-tasks from artefact-tasks. A short list of cognitive tasks you commit to performing without the model — the ones whose value is in the thinking, not the output.
2. Sit with a question for ten minutes before reaching for the model. Most thinking happens between minute three and minute eight of unbothered attention. The reach-for arrives too early to let the body produce its own thought.
3. Write the first draft by hand. Pen and paper, or an editor without the model. The first draft is where most of the synthesis lives. Subsequent drafts can use the model without sacrificing the cognitive deposit.
4. Re-derive a recent model-assisted argument. Take a piece the model helped you produce and rebuild the argument from memory. The re-derivation reveals what was thought and what was merely typed.
5. Notice the discomfort with unanswered questions. When you reach for the model, ask whether the question had been allowed to be unanswered. The reach-pattern is data; the body is calling for relief from a state that was where the thinking would have happened.

Reflection questions

• Which cognitive tasks have you stopped performing without the model in the last six months?
• When you read something you produced with heavy model assistance, can you defend the argument in a conversation?
• What does it feel like, now, to sit with a hard question for ten minutes without reaching for help?
• Which thinker-tasks, if you protected them deliberately, would do most to rebuild your sense of being able to think?

Frequently Asked Questions