Wrapper

Every AI product that calls an external model API is technically wrapping that API. The question is how much value the wrapper adds. A wrapper that simply relays your message to GPT-5 and shows you the answer is a thin wrapper: the model is doing all the work, there's no proprietary logic, and the product would evaporate if the underlying model improved enough or became available more cheaply elsewhere.

The critique of 'just a wrapper' became common in 2023-2024 as thousands of products launched on top of GPT-4. The concern: if your entire moat is a nice interface to a commodity model, what happens when the model provider builds the same interface, the model gets much better so the interface matters less, or a competitor makes the same wrapper for free? Many wrapper businesses did struggle as model capabilities improved.

The counterpoint: 'wrapper' with enough proprietary data, fine-tuning, deeply integrated workflows, or a strong distribution advantage is a real business. The word describes an architecture, not a destiny. Products like Notion AI, Intercom Fin, and Harvey are wrappers in some sense, but with deep workflow integration and domain data that create genuine differentiation. The question is always: what does your wrapper add that the model itself can't?