AI Wrapper

"It's just a wrapper" is one of the most common insults aimed at AI products, and it's sometimes fair and sometimes not. When the criticism is fair: the product's only real feature is a system prompt and a chat window a user could roughly replicate themselves in an afternoon with their own API key. No proprietary data, no workflow integration, no distribution advantage beyond having built the UI first. That product has no defense against a competitor, or the model provider itself, shipping the same thing as a built-in feature. When the criticism is reflexive and unfair: leveled at any product that happens to call an LLM API, as if using someone else's infrastructure were inherently illegitimate. Plenty of durable software is built on infrastructure the company doesn't own: Stripe wraps card networks, Zapier wraps other companies' APIs, and nobody calls those products "just wrappers," because the value sits in the workflow, the data, and the distribution, not in owning the layer underneath. How to tell which one you're building: would the product survive if a competitor could call the exact same model API tomorrow? What's the moat once the underlying model commoditizes further, proprietary data the model gets grounded in, a workflow users are already locked into, integrations that took real engineering, or distribution a competitor can't easily match? If the honest answer is "nothing beyond the prompt," the criticism is worth taking seriously before investing further, not dismissing as haters online.

A wave of AI meeting-notetaker apps in 2023 got labeled wrappers because their only feature was Whisper transcription plus a GPT summary, easy to replicate in a weekend. A support tool built on the same underlying model avoided the label because its real value was years of a company's own ticket history the model was grounded in, something a competitor couldn't copy just by calling the same API.