Decision Outsourcing to AI

A simple explanation

You open the chat window before you open the document. Help me decide what to write. Help me pick a title. Help me choose between these two options. The model responds. You take its answer, sometimes with a small adjustment, and the choice is made. The friction is gone. The choice is also gone — in the sense that you did not, this time, sit with it and feel which way you actually lean.

This is decision outsourcing to AI. Not the use of a tool to think alongside you. The handover of the deciding itself, so that the part of you that would have decided does not get the reps.

An everyday example

It is Tuesday evening. You have to decide between two job offers. They are roughly comparable on salary, very different on shape. You have known this is your decision for nine days. You have not, in those nine days, sat with either offer for more than four minutes before getting up. Tonight you open the chat window. Help me think through this. You paste both offers. You answer the model's clarifying questions. Forty minutes later you have a recommendation, three pros and cons per offer, and a frame for what to ask each company.

You accept the recommendation. You sleep. You wake up faintly relieved and faintly unsure whether the relief is from having decided or from having been told. By Friday the unsure feeling has a name: I do not know if I chose this or if it was chosen for me. The deposit — a sharpened internal model of what you actually want from work — did not land.

Why do I ask AI before I even ask myself?

Because asking the model is cheaper than asking yourself, and the Reward System is a cost minimiser. Sitting with a choice asks the body to hold uncertainty, to feel the pull of both options, to register the small grief of the path not taken. Asking the model asks the body to type. The trade looks rational in the next ninety seconds. It looks different across months.

There is also a subtler driver. The model never refuses. It never says I do not know what you want. It always produces a structured answer. For a system that has spent years experiencing its own decisions as effortful and uncertain, a faculty that returns confident structure on demand is genuinely consoling. The System reads the consolation as evidence the tool is the right instrument.

The behavioral loop

The loop that hides because it looks like productivity:

1. Trigger — a choice arrives. Small (what to eat, what to title an email) or large (a job, a move, a relationship pivot).
2. Friction registration — the body notices that deciding will require sitting with uncertainty. A small downshift, a small avoidance.
3. System re-route — the Reward System routes to the path with the lowest immediate cost: ask the model.
4. Prompt construction — you type the question, sometimes with care, often with a felt sense of relief that the work has begun.
5. Answer arrival — the model returns a structured response. The relief of structure lands as a clean reward signal.
6. Adoption — you adopt the answer, sometimes with a small adjustment to feel like a contribution.
7. Brief efficacy — the choice is logged. The System reads it as a win.
8. Residue — across days and weeks, the felt sense accumulates: I cannot tell what I think. The deciding faculty has been getting fewer reps.

Emotional drivers

The drivers are not laziness. They are usually three:

• A history of decisions that felt punitive — second-guessing, post-decision regret, social cost — which trained the body to read deciding as dangerous.
• A genuine appreciation for the tool's competence, which is real and which makes the substitution legible as good practice.
• A quiet, often unnamed wish to be relieved of the responsibility for outcomes — if the model said it, the loss is partly the model's.

The third one is rarely conscious and is usually the load-bearing driver.

What your nervous system does

A choice that you sit with produces a slow oscillation in the body — a back-and-forth weighting of options, accompanied by small somatic markers (Damasio's somatic-marker hypothesis). The oscillation is the decision happening. It is not pleasant; it is also not pathological. It is the felt shape of weighing.

When you outsource, the oscillation is cut short. The relief is immediate and reads as parasympathetic settling. But the body did not finish the weighing — it just stopped. Over time, the somatic markers themselves grow quieter, because the system has learned that the markers will not be used. The body that once told you this one in a slow inward turn now offers a flatter signal, because the signal has not been the thing the system was listening to.

The DojoWell interpretation

Decision outsourcing to AI is a textbook substitution in the MDT sense. The Reward System's ask was efficiency — less friction, faster resolution, more bandwidth for other things. The substitute it supplied was delegated cognition that wears the shape of efficiency. The substitute and the original share a surface property: in both cases, a choice gets made and you move on. They diverge on the inside.

A decision you make updates an internal model of how I weigh. The next similar decision arrives with a slightly sharper instrument. A decision you delegate makes no such update. The choice is logged externally, the internal model stays where it was, and the next similar decision arrives with the same dull instrument and the same friction. The System, encountering the same friction, re-routes again.

The density verdict is low not because the tool is bad but because the loop denies the deciding faculty the iteration that would have made the next decision lighter. False progress is the right signature: each delegation feels like a clean win and is logged as one, while the underlying capacity slowly erodes.

The work is not to stop using the tool. It is to use it as a thinking partner — one that surfaces frames, tests counter-arguments, summarises trade-offs — while keeping the actual I choose this one operation inside you.

How do I keep AI as a thinking partner without making it the decider?

A workable line:

1. Use it for structure, not for verdicts. Ask it to map the trade-offs, surface what you may be missing, steelman the option you are resisting. Do not ask it which to pick.
2. Decide in your own voice. After the model has helped, close the window. Sit for ninety seconds. Say internally, in plain language, which one you are choosing and why. The said-aloud verdict is the rep.
3. Notice the friction you were avoiding. If you reached for the model the moment you noticed uncertainty, that uncertainty was the work. The model can help you think about it; it cannot do it for you.

Practical steps

1. Track one week of delegations. A small log: what you outsourced, what the model said, what you did. Patterns become visible inside three days.
2. Reserve a class of decisions for yourself. Pick a category — what to wear, what to write, what to say to one specific person — and keep the model out of it. The reps matter more than the optimisation.
3. Ask the model to refuse to decide for you. Prompt: do not give me a recommendation; surface what I have not considered. The constraint reshapes the output and protects the deciding faculty.
4. Run a slow decision once a month. Pick something genuinely yours. Decide it without the model, over several days, with a notebook. The point is to remember what the deciding faculty feels like when it is allowed to run.
5. Repair the third driver. If the load-bearing reason for outsourcing is wanting to share the responsibility for outcomes, the work is upstream of the tool. Name it. The naming changes what the tool is being asked to carry.

Reflection questions

• Which class of decisions have you stopped making in your own voice?
• When you delegate, what is the relief actually relieving — the friction of the choice, or the responsibility for the outcome?
• Where in your life has the felt sense of I cannot tell what I think begun to accumulate?
• What would a healthy thinking-partner relationship with the model look like, specifically — what does it do, what do you do?

Frequently Asked Questions