"It's just a wrapper" has become the easiest way to dismiss an AI startup in a partner meeting. Sometimes that's fair. Sometimes it's lazy, and the fund misses a great company.
The real question isn't whether a startup uses someone else's model. Almost all of them do. It's whether anything about the business would survive if a competitor, or the model provider itself, decided to build the same thing next quarter.
Signs you may be looking at a wrapper
- The demo is the product. Remove the model and very little remains.
- One provider, no fallback. Every request goes to a single external model.
- No data advantage. The product doesn't collect anything a competitor couldn't get.
- Prompts are the main IP. Founders talk about their prompts the way others talk about their patents.
- Margins move with model prices. When the provider changes prices, the company's gross margin changes with it.
Questions to ask founders
- What would a competitor need to copy you, and how long would it take?
- Which models do you use, and what happens if your main provider doubles its price?
- What data do you collect that improves the product over time?
- How much of your value would remain if the model became free for everyone?
- What does it cost to serve one customer, and how does that change at scale?
When a wrapper can still win
Not every wrapper is a bad investment. Some win on distribution, a deep workflow in a specific industry, or a brand customers trust. The question is whether the company is building one of those advantages fast enough, before the underlying model makes its features standard.
The fastest way to check
AI Due Diligence scores "real AI or wrapper" and "AI models used" for any startup in about 10 minutes, so you can ask sharper questions in your next founder meeting.