Notes from an AI lab building agent swarms that model human behavior. Guides and research on synthetic users, synthetic user research, digital twins, AI user research, and behavioral science, for UX researchers, designers, and product teams deciding what to ship.
Start here
A plain definition, how a synthetic study actually runs, what the peer-reviewed evidence supports, and the questions it still cannot answer. Written for teams evaluating the category for the first time.
Perspectives5 min read
How-to6 min read
Ten questions that separate a grounded simulation platform from a prompt with a dashboard, why each one matters, and what a good answer sounds like. Includes the questions we would want asked of us.
Perspectives6 min read
The three ways a product team gets an answer about a design, what each one actually measures, what each costs in calendar time, and a rule for picking between them on the decision in front of you.
Perspectives8 min read
Foundation models are becoming a commodity input for user research. What does not commoditize is a panel grounded in your evidence, scored against decisions you shipped, and aimed at questions only you can invent.
How-to3 min read
The fastest way to judge synthetic research is to read one report. A plain walk through the sample simulation that ships in every new Sentia workspace, and how to read any report like it.
Research notes4 min read
In Nature, GPT-4 simulations predicted the results of 70 preregistered experiments at r = 0.85, and r = 0.90 on unpublished studies. What this validates, what it does not, and what to do with it.
How-to5 min read
A day-by-day itinerary from empty workspace to a grounded panel you have personally backtested, with what to expect at each step and what not to trust yet.
How-to6 min read
Simulation quality is mostly a grounding problem. A practical guide to feeding synthetic panels the transcripts, tickets, and behavioral data that make their answers worth acting on.
How-to5 min read
A working split for teams adopting simulated research. Which decisions can move on synthetic evidence alone, which need real users, and how to run the loop between them.
Research notes4 min read
Persona documents summarize research that already happened. Reasoning agents let you keep asking. What changed technically, what the peer-reviewed evidence shows, and what still does not work.
Perspectives4 min read
When a directional study costs an hour instead of six weeks, the constraint on product judgment moves. What changes for teams, what happens to researchers, and what expensive research is still for.