Questions
Questions about AI that knows how you decide
Eleven questions, each answered on its own page. The answer is in the first sentence of each. Nothing here is gated and nothing here is for sale.
These are the questions that come up once someone has been using AI seriously for a while and has noticed that the problem is no longer capability. The tool does competent work. It keeps doing competent work that is not quite the work they would have done, and no amount of better prompting closes the last piece.
Each page answers one question and stands alone. Where a question was already answered properly elsewhere on this site, the page was cut rather than duplicated.
The questions
Can AI learn how I make decisions?
Told, not learned
It can hold how you decide. It cannot derive it. The difference only shows up on a case you have never handled before, which is also the only kind of case where it matters.
Why doesn't ChatGPT remember how I think?
The eviction policy
It remembers what you volunteered, and what determines your quality is never volunteered. Worse, memory has a budget, and the entries it drops first are the rare ones, which are the valuable ones.
How do you train AI on your own judgment?
Interception, not a project
Not by sitting down to write it out. By intercepting the corrections you are already making, in a four-part shape, of which the part almost everyone omits is the boundary.
Can you upload your judgment to an AI?
Nothing is in a file
No, because there is no stored copy to move. Judgment is rebuilt every time you use it. What can happen is elicitation, which is a rate rather than an event.
Why does my AI output feel generic?
Correct for the question asked
Because generic is the right answer to what you asked. It is only generic relative to a standard you did not supply, and style instructions cannot fix it, because it is a decision problem wearing prose clothes.
Is an AI clone of yourself worth it?
Fronting against deciding
Yes if its job is to be seen. No if its job is to decide. One buying rule separates them, and it is about what a confident wrong answer costs you.
Why does AI get you 80% of the way there?
It cannot see that it stopped
Because the part it produces is the part already written down somewhere by someone, and the rest exists only inside your business. The real problem is that it delivers the eighty with the confidence of the hundred.
Can AI replace a founder's judgment?
Dilution, not replacement
No, and replacement was never the risk. The risk is dilution: your judgment leaves the business one defensible automated decision at a time, and unlike replacement, nobody sees it happen.
What should a founder automate first?
Sort by what a mistake costs
Not the repetitive work. Sort every task by what a wrong output costs. Automate what costs money. Capture, before you automate, anything that costs trust, and never automate the function you win on before capturing what makes it win.
How is an imprint different from a prompt library?
Indexed by task against indexed by case
A library is indexed by task, so retrieval is your job and you only reach for what you remember owning. An imprint is indexed by case and arrives without being asked for. And a library holds what you want done, never what you would refuse.
What happens when AI guesses your preferences?
It regresses you to your own mean
It infers from your most frequent behaviour, and your frequent behaviour includes every compromise you made under time pressure. So it does not learn your standard. It learns the version of you that was tired.
If you would rather start with the definitions
The category itself is defined at what Imprinted AI is. The vocabulary these answers use is at the glossary, fifteen terms with an argument each. Where this approach sits against the alternatives, including the several cases where an alternative is the right tool, is at the comparisons. Original material from actually running one of these is in the field notes.
One page that is not here
"What is tacit knowledge in business" was planned as a twelfth question and cut. It is answered at tacit knowledge in business, which carries the term definition as well, and two pages of ours competing for one question is worse for a reader than one good page.
Keep going
These eleven answers are the parts of a single argument, taken separately. The argument itself is longer and runs in one direction.
The full case is The A.I. Business Manifesto. About 23,000 words, free to read on the page, no gate in front of it. If you would rather have the short version, the same page will send you the three fixes and the PDF.
Read The A.I. Business Manifesto
Free either way. Reading it costs nothing and asks nothing.
Last updated: 28 July 2026