Every AI startup idea meets the same objection: a vendor will ship this as a feature and there will be nothing left. It is a fair question. It is also not the one that does the killing. On the single Discovery run we publish in full, the check that asks it was named on 6 of the 36 ideas the run dropped — while the check asking whether there was a documented problem behind the idea at all was named on 21.
What the run actually dropped
The run is one sweep, published unedited with every rejection and its reason attached, which is the only reason this page can put a number on any of it. 69 ideas entered the funnel and 36 were dropped. Each drop carries the checks it failed.
Two of those checks matter here, and the distance between them is the finding.
Both bars are drops from the same run, scaled to the larger of the two and counted by the same
function /sample-discovery renders — so the chart cannot disagree with the page it
came from. The missing problem outkilled the wrapper question 3.5 to one.
This is not an argument that the wrapper question is soft. It is an argument about order. If you are holding an AI idea and worrying about a vendor roadmap, you are worrying about a question far down the list while the first one on it goes unexamined.
The wrapper objection almost never arrives alone
The stronger detail is in how those 6 drops were composed. Only 1 of them died of the wrapper question by itself. The other 5 each failed something else in the same pass.
One row per dropped idea the AI-dependency check was named on, with whatever else the same idea failed beside it. Count the rows that end at "nothing else" — the wrapper problem is usually a symptom of something that already has its own name.
What accompanied it, on this run:
Too much hands-on support (2), We found no documented problem (2), Needs heavy compliance (licenses, certifications) (1).
Read that list back and it stops sounding like an AI problem. An idea that is a thin layer over a model is also, usually, an idea nobody documented a problem for, or one that needs a human in the loop for every customer, or one that cannot launch without a compliance review. The wrapper is the visible symptom because it is the part you can see from outside.
What to check before you worry about the wrapper
The cheap checks go first, and this is the practical consequence of the two figures above. Before asking whether a vendor will absorb your product, ask:
- Is there a documented problem behind it? Not a plausible one — one you can point at, on a page somebody else wrote, with a link that opens. If you cannot produce that link in an afternoon, the vendor roadmap is not your problem. Where to find customer problems is the method we use.
- Can it run without you in the loop? A model that needs supervision on every output is a support load wearing a product's clothes, and support load was the most common companion above.
- Can it launch without a licence or an audit? The compliance check killed alongside the wrapper check here, and it is the one people discover latest.
Each of those is settled by reading rather than building. The wrapper question, by contrast, is a forecast — and a forecast is the most expensive kind of question to answer first.
Where the AI question does bite
It bites when the answer is genuinely nothing. If the whole product is convenient access to a capability somebody else owns, then the thing keeping you alive is their inattention, and pricing that as durable is how founders get surprised by a changelog. That is a real failure mode with a real taxonomy behind it — accumulated data, workflow depth, integration surface, distribution, accountability — and it has its own page: tech startup ideas and the question that now decides them.
The point here is narrower. That question is one gate among the ones a candidate has to clear, and on the run we can actually count, it was not where most ideas fell over.
What this does not prove
One run, one market, one day. Three limits worth stating before you lean on any of it:
- It is a sample of our sourcing, not of AI ideas in general. The sweep read public pages in the markets it planned. A different market would return different counts, and nothing here claims a rate.
- The judgements underneath are model output. They sample fresh each run, so a second sweep would not reproduce this list. The arithmetic on top of them is fixed; the ratings are not.
- These are our checks. An idea dropped for leaning on one vendor was dropped because a solo founder in that position is exposed, not because nobody would buy it. Read them and disagree where you want to.
The short version
- The wrapper objection is the one everybody raises about an AI idea, and on the run we publish it was named on 6 drops against 21 for the missing problem.
- Only 1 of those drops was the wrapper question on its own. It usually arrives with a support, compliance or evidence problem attached.
- Ask the cheap questions first. A documented problem is an afternoon of reading; a vendor roadmap is a forecast.
- When the honest answer to "what is left if they ship it" is nothing, that is a real kill — just not the common one.