Tech • AI • Robotics • Game
Sites and the Next Generation of Web Apps | DevDay 2026
9.2/10TL;DR
OpenAI said Sites, its web app building and hosting platform, has grown to 8 million hosted sites and is being used internally by more than 70% of employees to turn documents, spreadsheets and ideas into shareable, agent-ready applications.
KEY POINTS
Rapid internal adoption
Sites began as an easier way to deploy internal full-stack tools and quickly spread beyond engineering teams. OpenAI said more than 70% of employees now build their own sites, using them for prototypes, internal tools, interactive presentations, team dashboards and experiments. The company described the result as a workplace where there is effectively a site for every use case.
Scale just months after launch
Although Sites launched only a few months ago, the platform is already hosting 8 million sites. The service is positioned not just as basic web hosting, but as a system for generating, publishing and maintaining full-stack apps directly from prompts through Codex.
Apps built from a single prompt
In one example, a launch-tracking spreadsheet was turned into a “launch radar” app by asking Codex to create a new site with list, board and calendar views, milestone tracking and direct links to project documents and assets. The app was built within minutes and used a shared Devday design library, showing how teams can convert static internal documents into interactive interfaces without manual coding.
Automatic updates through plugins and scheduled tasks
OpenAI introduced scheduled tasks for sites, allowing apps to refresh themselves hourly, daily or weekly by pulling data from connected services. The system combines plugins with scheduled jobs and a built-in SQLite database, so new data can update the app instantly without redeploying source code. In the demonstration, a site was configured to refresh every morning using launch briefs stored in Google Drive.
Built-in authentication and private sharing
New sites are private by default and use Sign in with ChatGPT for access control. Public publishing is available, but private deployment is designed to ensure only intended users can enter a site. Creators can also invite specific colleagues and grant editor permissions so others in the workspace can push updates to the hosted app.
Personalized views using each user’s connectors
A new capability lets users who open a private site bring their own ChatGPT plugins into that site. Services such as Drive, Slack and Linear can then supply user-specific data at load time. In practice, two people viewing the same launch app saw different Linear tickets assigned to them, without the site developer needing to build custom third-party integrations, manage API keys or handle separate permission logic.
Read and write actions inside connected apps
The same connector model supports actions as well as data retrieval. In the demonstration, a user changed the status of a Linear ticket from inside the site and saved the update back to Linear. The feature is intended to make sites function like collaborative operational tools rather than static dashboards.
Editing through annotations
OpenAI also introduced annotation extensibility in the ChatGPT in-app browser. Developers can embed custom annotations directly into a site’s HTML so editors can select interface elements, see suggested prompts and discover available changes more easily. The company said this is especially useful for design systems and shared component libraries because it lowers the barrier for non-developers to modify an app safely.
Agent-friendly apps with WebMCP
Another major addition is WebMCP, described as a browser standard that lets agents interact with sites through exposed tools instead of clicking through interfaces. That approach is meant to be faster and cheaper than general computer-use methods because agents can query app backends directly. In one consumer-style example, an app used Slack context and site tools to recommend gift items, update the page live and generate a personalized message, all with the user collaborating alongside the agent in the browser.
CONCLUSION
OpenAI is positioning Sites as a low-friction way to build and host full-stack apps that are continuously updated, identity-aware and usable by both humans and software agents. The broader bet is that web applications will increasingly be designed not only for direct users, but also for the agents acting on their behalf.
Ask a question