Wrapper or Winner

Every AI product faces one existential question: are you a wrapper that gets squeezed when the model layer moves, or a winner that compounds regardless?

The distinction

Wrappers add UI on top of an API. Their value proposition is convenience — they make a foundation model easier to use for a specific task. When the model provider ships that same convenience natively, the wrapper dies. Jasper, Copy.ai, and dozens of "AI writing assistant" startups learned this when ChatGPT shipped custom GPTs.

Winners own something the model can't replicate:

Distribution — they already have the customers (HubSpot adding AI to an existing CRM)
Proprietary data — their dataset compounds with usage (Bloomberg Terminal, Palantir)
Workflow lock-in — switching costs are structural, not emotional (Salesforce, Figma)
Domain expertise — the model is a component, not the product (Veeva in pharma, Toast in restaurants)

The stress test

Three questions to ask about any AI product:

If the foundation model adds this feature natively, do customers still need you? If no, you're a wrapper.
Does your product get better with more usage? If no, you have no data moat.
Would a customer need to rebuild their workflow to leave? If no, you have no switching costs.

If all three answers are no, you are a wrapper. This is not a death sentence — it's a signal to pivot before the model layer catches up.

Why this connects to obligation marketing

Winners build obligation-marketing naturally — their products create process lock-in (rung 6) and expertise obligation (type 4 in the obligation-marketing-taxonomy). Wrappers almost never reach rung 3.