Developers just got a direct pipeline into ChatGPT. OpenAI’s DevDay, held October 6, 2025, brought AgentKit—a tool that lets coders build applications not alongside, but inside the chat interface itself. The company, founded in 2015 in Delaware as a nonprofit, is betting that the next wave of AI adoption will come from turning its flagship product into a platform.
ChatGPT launched in November 2022. It catalyzed the AI boom. Millions of users poured in. But for all its power, the chatbot was a closed box—you talked to it, it answered, and that was the loop. AgentKit breaks that loop. Developers can now embed custom logic, data sources, and entire workflows directly into the conversation. The box is no longer sealed.
This matters because of what it unlocks. OpenAI has already produced the GPT family of large language models, the DALL-E series of text-to-image models, and the Sora series of text-to-video models. Those are powerful engines. But they have mostly run on OpenAI’s servers, behind OpenAI’s interfaces. AgentKit shifts the power to the developer’s hands. An accountant could build a tax-prep app that lives inside ChatGPT. A doctor could wire in a diagnostic tool. A teacher could create a lesson planner that pulls from a school’s own database. The model becomes the operating system, not just the application.
The stakes are concrete. For developers, the risk is dependence on a single platform—build inside ChatGPT, and your app lives or dies by OpenAI’s rules, pricing, and uptime. For OpenAI, the risk is losing control of its own creation. The company’s for-profit subsidiary, created to support the nonprofit’s mission, has driven commercial innovation. AgentKit is the most aggressive move yet to lock developers into its ecosystem. If it works, ChatGPT becomes the default environment for building AI-powered services. If it fails, developers walk away with nothing but wasted time.
For users, the stakes are simpler. They get more capable assistants. Instead of a chatbot that can only talk, they get one that can act—file forms, query databases, generate reports. The line between talking to an AI and using software dissolves. That is the promise. The peril is that the AI becomes a gatekeeper, mediating every interaction through a single corporate interface.
OpenAI has been here before. The release of ChatGPT in 2022 brought its research to a wider audience and sparked generative AI’s mainstream moment. Since then, the company has pushed boundaries. AgentKit is the latest push. It is not a new model. It is a new way to use old models. That distinction is easy to miss but hard to overstate. Models are tools. Platforms are ecosystems. AgentKit is the bridge between the two.
Developers at DevDay got the first look. They saw a system that lets them build without leaving the chat window. No separate server. No separate front end. No separate deployment pipeline. Just code that runs inside the conversation. It is a radical simplification of the development process. It is also a radical concentration of power in one company’s hands.
OpenAI, founded in 2015 in Delaware, has always had a dual identity: a nonprofit mission to advance AI safely, and a for-profit arm to fund that mission. AgentKit serves both. It advances the technology by making it more accessible. It advances the business by making the platform stickier. Whether that balance holds depends on how developers use the tool—and whether they trust the hand that built it.
