The AI Gold Rush
For the last few years, building a startup on top of OpenAI's API has been one of the hottest tickets in tech. The appeal is obvious: instead of spending millions on foundational research, a small team can tap into a world-class large language model for
a relatively low cost. This sparked a Cambrian explosion of 'wrapper' startups—companies that put a user-friendly interface on top of OpenAI's powerful engine to solve a specific problem, from writing marketing copy to summarizing legal documents. The strategy was simple: find a niche, build a slick workflow, and ride the AI wave. Investors poured money into this ecosystem, and a vibrant community of developers pushed the boundaries of what was possible with just an API key. For a time, it seemed like a perfect symbiotic relationship.
The Platform Giveth, and It Taketh Away
The problem with building your house on someone else’s land is that the landlord can change the rules at any time. In the tech world, this is often called 'Sherlocking'—when a platform owner (like Apple or, in this case, OpenAI) sees a popular third-party application and builds its core functionality directly into the native platform. Suddenly, a feature that was once a startup's entire value proposition becomes a free, built-in part of the tool everyone is already using. OpenAI's DevDays have become a prime example of this phenomenon. With each event, the company releases more powerful, cheaper, and more capable tools. What was once a complex task requiring a specialized 'wrapper' might now be achievable with a simple prompt in the base ChatGPT interface or a new, more advanced API call. Announcements of native PDF reading, advanced data analysis, or agent-like capabilities can render entire categories of startups obsolete overnight.
Redefining the Moat in an AI World
So, is building on OpenAI a fool's errand? Not necessarily, but the rules of the game have changed. Defensibility—a startup's ability to maintain a competitive advantage—can no longer come from simply being a thin layer on top of an API. The new moats are deeper and harder to build. One key area is proprietary data. A startup that can leverage unique, hard-to-replicate datasets can produce better results than a generic model. Another is deep workflow integration. Instead of just being a tool, a startup can become an indispensable part of a company’s specific, high-stakes operational process, creating high switching costs. Focusing on a niche vertical, like a specific area of law or medicine, allows for the development of highly specialized solutions that a generalist platform like OpenAI is unlikely to pursue. Finally, brand and user experience still matter immensely. A product that is a delight to use and builds a loyal community can create a defense that technology alone cannot replicate.
