How to build an API-centric AI platform
OpenAI’s approach with their plugins strategy is a lesson to every founder trying to build a platform. OpenAI announced their ChatGPT plugins offer on March 23, 2023. Their announcement blog post says they “are gradually rolling out plugins in ChatGPT so we can study their real-world use, impact, and safety and alignment challenges.” They’re launching the plugin platform with several third parties, including Wolfram and Zapier.
Let’s see how OpenAI has evolved to reach the point where it’s now becoming a platform. It started as something that you could “play with” to chat with a bot that would answer questions, remember? OpenAI launched ChatGPT in November 2022 to get its AI engine in front of as many people as possible. By launching something far from perfect, they started gaining knowledge about what people really wanted from an AI-powered chat system. As more people were using the chat system, OpenAI evolved it by fixing bugs, making it faster, adding layers of feedback loops, and, most importantly, learning their “product holes” (more on that later).
Despite being incorrect in many situations, the chat product got to a point where demand was so high that OpenAI launched a paid version in February 2023. ChatGPT Plus offers high availability, faster response, and early access to new features. This was a way for OpenAI to introduce rate limiting into the system. As demand grew and new use cases appeared, OpenAI introduced new and better AI engines. At the same time, usage of the AI engine through their API increased with third parties launching features and products backed by OpenAI’s engine.
OpenAI could have stopped here and used the chat UI and the API as revenue drivers. After all, they already have two monetization models in place. ChatGPT is a subscription-based service, and OpenAI’s API uses the pay-as-you-go model. Instead, they analyzed the holes that their offering has. Instead of building more features to fill those holes, they decided to open the chat UI to third parties by launching a plugin system. In other words, they have created an AI platform where anyone can plug into their engine to augment it with external capabilities.
What’s even more interesting is how OpenAI is building its plugin ecosystem. The foundation of a plugin is actually an OpenAPI definition. OpenAI’s engine understands what the OpenAPI document defines and makes those operations available at runtime. In a situation like this, and if OpenAI grows a lot, we’ll get to a point where everyone will want to have an API so they can be part of the ecosystem. We’ll transition into a stage where API products will be focused on single features so chat users can easily consume them.
In these times, thinking about APIs as products is critical. Being able to design APIs following a product mindset is necessary to play in this platform-centric ecosystem.