Everyone building on a model gets the same objection: it is just a wrapper, a vendor will ship this as a feature.

I counted how often that was actually the cause of death on a run I publish in full, and it was not where the ideas fell over.

69 candidates, 36 dropped. The wrapper check was named on 6 of them. The missing-documented-problem check was named on 21. That is 3.5 to one.

I am not arguing the wrapper question is soft. I am arguing about order: if you are worrying about a vendor roadmap before you have found a single written complaint the idea would answer, you are working the wrong end of the list.

The detail I found more useful than the ratio. Of those 6 wrapper drops, exactly 1 died of the wrapper question alone. The other 5 failed something else in the same pass — support load, a missing problem, a compliance barrier.

Which reframes the objection. A thin layer over a model is usually ALSO an idea nobody documented a problem for, or one that needs you in the loop for every customer, or one that cannot launch without an audit. The wrapper is just the symptom visible from outside.

So the practical order is the cheap questions first, because all three are settled by reading. Is there a documented problem, with a link that opens. Can it run without you in the loop. Can it launch without a licence. The wrapper question is a forecast, and forecasts are the most expensive thing to answer first.

Where it does bite: when the honest answer to what is left if they ship it is nothing. If the product is convenient access to a capability somebody else owns, your moat is their inattention. That is a real kill — just not the common one.

One run, one set of markets, my checks, no rate claimed: https://whittleos.com/guides/ai-startup-ideas