AI should adapt to you, not the average of everyone.
Personalized AI that learns how you want things done — not just what you want.
Idios learns each person's procedural preferences — how much control they want, when an agent should ask, and what they consider a good outcome — and makes agents behave accordingly.
Idios · Greek for one's own
Every person generates thousands of signals about how they want things done. Today, all of them are averaged away.
People differ in ways that matter to every task an agent touches: risk tolerance, desired level of control, speed versus thoroughness, privacy expectations — even what counts as a satisfactory outcome. Today's AI compresses that diversity into one universal optimal behavior, trained on averaged preferences.
So the same action can delight one person — it handled everything — and destroy another's trust — it acted without asking. Everyone gets the same agent, and humans bend themselves around it.
There is a distinction hiding here that nobody is collecting. Today's preference data measures what an agent produces: which output is better. That is outcome taste, and it converges toward one shared answer. Procedural preference is a different axis entirely — how the work should be done, when to ask versus act, how much to verify, what to do when something is ambiguous. It diverges by person. And it is exactly what agents acting on your behalf actually need.
It arrives now for a reason. Correctness came first: can the agent do the task at all? Outcome taste came second: is the result any good? Procedural preference is third, and agents only just crossed the first two thresholds.
Human diversity should be represented inside AI systems, not averaged away.
Agents built this way take over the parts of work people genuinely want to delegate — while preserving control, individuality, and human agency.
Behavior is the signal
What people actually do — correct, override, take control, redo, abandon — says more about how they want to be served than anything they would write in a settings panel.
Difference is not noise
The variance between people is the most valuable thing in the data. Averaging it produces an agent that is acceptable to everyone and right for no one.
Preferences belong to you
The long-term product is a portable memory of how you want work done — owned by you, and carried into any new agent, product, or robot so it immediately knows how to work with you.
Notes on the thesis
Essays on procedural preference, why trust breaks between a person and an agent, and what personalization has to become as agents take on consequential work.
Leave a note and I'll send them when they're out.