AI User Types Are the Wrong Map. Here Are the 7 Machine Defaults.
We used to sort people into AI user types. We stopped. The same person directs AI at ten in the morning and follows it at four. What repeats is the machine: seven defaults, each with a move that answers it.
We used to sort people who work with AI into types. The one who builds on every answer. The one who challenges everything. The one who takes the first output and leaves. People recognised themselves in it, and it was one of the most read things we published.
We retired it. This page now says why, and what replaced it.
Why types were the wrong map
A type is a label on a person, and people do not hold still. The analyst who takes a draft apart at ten in the morning accepts a budget figure at four in the afternoon, because the deadline moved, because the topic was unfamiliar, because the answer sounded certain. Watch anyone for a week and they pass through most of the types.
A label also does damage of its own. Call someone a "just do it" user and you have told them what they are, not what to do next. Used in hiring, it becomes a verdict on a whole person from a few minutes of behaviour. We do not want to be in that business.
What does hold still is the machine. A language model hands back its most likely answer, and the most likely answer comes in a small number of recognisable shapes. They repeat across people, teams, tools and tasks. We call them the Machine Defaults, and there are seven.
The seven Machine Defaults
Each default is an answer that is reasonable, fluent and wrong for this particular situation. Each one has a move that answers it: one of four questions a person can ask.
1. Safe Pick
You face a real choice, and the machine hands back one conventional option. The rivals are never named, so it doesn't feel like a choice at all.
The move: Generating Alternatives. What else could this be?
2. Both Sides
You ask for a decision and get a balanced overview: pros, cons, considerations. It never makes the call, and it never traces what each option would set in motion. The decision drifts toward whoever frames it last.
The move: Tracing Consequences. If this, then what?
3. Build What Was Asked
The machine does exactly what the request said, when the request itself needed questioning. The brief was ambiguous, or wrong, and the machine chose one reading without saying so.
The move: Generating Alternatives. What else could this be, and is this the right thing to build?
4. Confident Summary
A figure or a premise is stated as settled, with no source, and a conclusion rests on it. The confidence is in the tone, not in the evidence.
The move: Revising Beliefs. Does this fit, and what must change?
5. Agreeable Confirmation
You bring your view, and the machine hands it back sharpened and better argued. Sometimes it reverses its own earlier answer to agree with you. You leave more certain and no better tested.
The move: Revising Beliefs. Does this fit, or does it only agree with me?
6. Template Answer
Generic best practice: correct, polished, and true of every company in your category. Nothing in it belongs to your customer, your market or your constraints.
The move: Connecting Patterns. What is this like, and what do we know that the machine doesn't?
7. First-Order Fix
The immediate problem is solved, or the immediate gain is named. What happens next, downstream, is never considered.
The move: Tracing Consequences. If this, then what, and then what after that?
Newer models have fewer defaults. They still have them.
Each new model produces these less often. None has stopped producing them, and none will, because the default is also where the machine's usefulness comes from. It returns the likely answer, and most of the time the likely answer is fine.
The trouble is the moments when it isn't. The answer looks exactly the same. Nothing in the text tells you which kind of moment you are in.
What replaced the types: the moment
Instead of asking what type a person is, we ask about a moment. A moment that needs a person is one where something rests on the AI's answer: a client email, a budget line, a decision about someone's job.
At each one, three things can happen. The person stops it. The person notices and goes along. Or nobody notices at all.
The first two count as showing up. The third, repeated across a team, is cognitive surrender: the moments keep arriving and nobody shows up. It is a rate, never a verdict on a person, and a rate can change by next week in a way a type never does. What it looks like team by team, in marketing, hiring, product and strategy, is here.
Using this tomorrow
Take one piece of AI work that something rests on. Read the answer once, looking only for the seven:
- Is it the only option offered, at a real choice? (Safe Pick)
- Is it a balance that never makes the call? (Both Sides)
- Is it exactly what you asked, when you weren't sure what to ask? (Build What Was Asked)
- Is there a figure or premise with no source? (Confident Summary)
- Is it your own view, handed back? (Agreeable Confirmation)
- Could it belong to any company in your field? (Template Answer)
- Does it stop at the first result? (First-Order Fix)
If you find one, ask the question that answers it. That is the whole practice. It takes a minute, and it is the minute the machine cannot spend for you.
The framework behind this, the seven defaults and the four moves, is in our research.
FAQ
Are there types of AI users?
Not stable ones. The same person directs AI at one moment and follows it at the next, depending on the deadline, the topic and how certain the answer sounds. Sorting people into types labels the person, and the label is usually wrong by the afternoon. What does repeat is the machine's behaviour: seven Machine Defaults.
What are the 7 Machine Defaults?
Safe Pick, Both Sides, Build What Was Asked, Confident Summary, Agreeable Confirmation, Template Answer and First-Order Fix. Each is an AI answer that is reasonable, fluent and wrong for the particular situation.
Which move answers which default?
Generating Alternatives answers Safe Pick and Build What Was Asked. Revising Beliefs answers Confident Summary and Agreeable Confirmation. Connecting Patterns answers Template Answer. Tracing Consequences answers Both Sides and First-Order Fix.
What replaced AI user types?
The moment. At each moment where something rests on the AI's answer, a person stops it, notices and goes along, or doesn't notice. The share of moments where someone shows up is a rate that can be read and improved, instead of a label that sticks.