Six years ago, OpenAI was a nonprofit research lab with an unusual mission: build artificial general intelligence and give it away. By November 2023, the company had become something else entirely. It held its first-ever developer conference, called DevDay, in San Francisco. The company announced GPT-4 Turbo, custom GPTs, and something called the Assistants API. This was not a research paper. This was a product launch.
The event marked a clear shift. OpenAI had spent years releasing models like GPT-3.5, then GPT-4. Those were foundation models — big, general-purpose language engines. Developers built things on top of them, but they had to do the heavy lifting themselves. The new tools changed that. Custom GPTs let anyone create a tailored version of ChatGPT without writing code. The Assistants API gave developers a way to build agents that could follow instructions, use tools, and retrieve information. The company was no longer just selling a model. It was selling an ecosystem.
GPT-4 Turbo itself was the headline. The original GPT-4 had been quietly powerful. An early version went into Bing Chat in February 2023. Then it hit ChatGPT in March. The model could generate text that read like a human wrote it. A variant called GPT-4V extended that ability to images — it could look at a picture and talk about it. But the company never revealed how big the model was. No parameter count. No training data size. That secrecy was typical. OpenAI had learned to keep its cards close.
The road to DevDay was not short. OpenAI started with GPT-1 in 2018, then GPT-2, then GPT-3, then GPT-3.5. Each version got bigger and more capable. Each version also got more expensive to run. The company had to change its business model to survive. It went from nonprofit to capped-profit. It started selling API access. It partnered with Microsoft, which poured billions into the company. By the time DevDay arrived, OpenAI was a major commercial force in AI.
The Assistants API was the piece that worried some developers. It gave OpenAI a way to host and manage AI agents inside its own infrastructure. Instead of building a chatbot from scratch and hosting it yourself, you could call OpenAI’s API and get a fully managed assistant. The trade-off was clear: convenience for control. OpenAI would own the stack. Developers would own the application layer.
Custom GPTs took that logic even further. They let users create a version of ChatGPT that knew specific things and behaved in specific ways. You could give it custom instructions, upload your own data, and share it with others. The company positioned this as a democratizing move. Anyone could build an AI assistant now. But the assistants lived inside OpenAI’s ecosystem. They were not portable. They were not open-source.
That tension ran through the entire event. OpenAI had started as an open research lab. Its founders believed in sharing everything. Over time, that philosophy eroded. The company stopped releasing full model details. It stopped open-sourcing its biggest models. It became a proprietary platform. DevDay was the logical endpoint of that evolution. The company had built a walled garden and was now inviting developers to plant inside it.
The industry noticed. Competitors like Anthropic and Meta were pushing open models. Startups were building on open-source alternatives. But OpenAI had something they did not: a massive user base, a working product, and a brand that meant “AI” to most people. DevDay was a statement. The company was not slowing down. It was building the infrastructure for a new kind of software platform — one where the intelligence came from the cloud, not the device.
None of this was accidental. The GPT-4 rollout had been careful, staged, and controlled. The Bing Chat integration in February 2023 gave Microsoft a marquee feature. The ChatGPT release in March gave consumers a direct interface. The API gave developers access. DevDay gave the whole thing a name and a future. The company was not just launching products. It was setting the terms for how AI would be built, sold, and used.
