Harvard and MIT just simulated 8.3 billion people, one AI persona for every human on Earth, and released the top 1 million of them for anyone to use.
1 million out of 8.3 billion. Do that math and odds are you didn't make the cut. Even in a simulation built to represent every human on Earth, most humans still got left on the cutting room floor.
Which is exactly the problem in UX research too. It's not sample size. It's not speed. It's compression.
Every research team does the same thing: run 20-30 interviews, then collapse them into 3-5 named archetypes because that's what a team can actually hold in their head and design against. "Persona 3: Busy Ben." Real users don't cluster that cleanly. The compression step throws away the contradictions and the long tail on purpose, just to make the data usable in a design review.
That's the actual cost of traditional personas. Not that they're wrong. That they're lossy.
A grounded, queryable research vault doesn't have that constraint. You're not forced to flatten your users into archetypes before you know what question you'll need to ask. Instead of "how would Busy Ben react to this," you can ask "how would power users in this segment who churned after the last pricing change react to X," a slice specific enough that it would never survive being generalized into a named persona in the first place.
That's the real shift. Not scale, resolution. The benefit isn't more personas, or 8 billion of them. It's not having to decide ahead of time which nuance is worth keeping.
The catch, and it's a real one: a synthetic query is still built on a model's priors. It can tell you what your past research implies. It can't tell you what your users would say about something nobody's asked them yet. That's still the job of an actual conversation with an actual person.
So here's the workflow I'd actually trust:
Query the vault at full resolution to generate the sharpest possible hypothesis.
Take that hypothesis to real users to find out where it's wrong.
Feed what you learned back into the vault.
Synthetic research doesn't replace the human. It just stops making you flatten them first.
Curious how other research and product teams are thinking about this. Are personas becoming obsolete as a format, or just as the only format?
#ProductManagement #UserResearch #AgenticAI #ArtificialIntelligence #ProductDesign