Isenberg runs 50 synthetic users through 30 days of his product

Greg Isenberg says he tests his own products with synthetic users, and he laid out the pipeline in one line. Fathom call recordings become LLM personas, the personas get browsers through Browserbase, and each one runs 30 simulated days inside the product.

GREG ISENBERG

@gregisenberg

X

the stuff they love on day 1 and hate on day 10 gets cut

GREG ISENBERG

@gregisenberg

X

i don't know why more people aren't doing synthetic user testing

The ratio he gives is one to one. Fifty customer calls become fifty personas, each with its own browser session, each spending a simulated month poking at the thing. The filter he applies at the end is the interesting part. Anything the personas love on day 1 and hate on day 10 gets cut.

Why the day 10 part matters

Day 1 reactions are cheap to collect. A landing page test, a demo call, a Product Hunt launch, all of it tells you whether something looks good. Retention is the expensive signal, because normally you have to wait an actual month with actual users before you find out which features quietly annoy people on the second week. For a solo founder that is a month of building on top of a decision you have not validated yet.

Isenberg is proposing a stand-in for that month. He is not claiming it replaces real users, and he does not post a result, a cost figure or a case where the simulation matched what live customers later did. What he says is that it is cheap and that he does not understand why more people are not doing it.

The framing

He opens with New Coke. In 1985 it beat Pepsi in 200,000 taste tests, then got pulled in 79 days. That is the whole argument compressed into one example. A test that measures the first sip will happily tell you to ship something people stop wanting by the third can.

Treat this as one operator's method rather than a proven technique. Synthetic personas inherit whatever the model thinks a customer sounds like, and nobody in the sources has published whether a simulated day 10 complaint predicts a real day 10 cancellation. The part worth stealing is the question Isenberg is asking, which is not whether people like it, but whether they still like it later.