Can AI Users Judge an App First?

How do you know whether the app you just built is any good—before real users tell you?

Before launch, you mostly have your own perspective. But you know the product too well. You know what every button is supposed to do, what every screen means, and why every design decision was made.

A real user knows none of that. They simply hesitate, get confused, lose interest, or leave.

So I built an AI agent skill called n-personas: a stable panel of diverse simulated users that tries the product before real users do.

The panel includes personas such as:

the impatient skepticthe anxious novicethe value comparatorthe distracted multitasker

Each persona runs as an independent AI subagent. It opens the product, walks through the actual flow, judges what appears on screen, and checks competitors when required.

Each one returns a direct verdict:

Will this user stay or bounce? Were their expectations missed, met, or exceeded? What created trust, confusion, interest, or friction?

The results are then synthesised into one product-level assessment: whether the experience is likely to attract, retain, and grow users—and the single highest-leverage improvement to make next.

A persona judges only the experience placed in front of it.

It gives no credit for the founder’s intent, the engineering effort, or how difficult the product was to build.

The system also defaults towards churn, because that is how the real world behaves: users are usually more likely to leave than stay, and conversion is generally much closer to zero than to certainty.

Positive feedback must therefore be earned through the actual experience. This avoids the common failure mode of AI reviewers that politely conclude everything is already good.

Before real users judge your product, let a tougher panel of simulated ones go first.

This introduces a new engineering question:

Can we create a new kind of persona-driven optimisation loop, in which an app or module repeatedly improves itself from persona feedback until the panel is satisfied?