News · OpenAI puts a hosted frontend behind Codex with Sites, annotations, and a product design plugin
OpenAI puts a hosted frontend behind Codex with Sites, annotations, and a product design plugin
A code tool that now generates shareable interactive webpages, edits them by pointed selection, and prototypes UI from a live URL
The partner list reads like a competitive map
OpenAI names early Sites partners including Vercel, Wix, Base44, Replit, Lovable, Figma, Webflow, and Emergent. That roster is notable because most of those names are themselves in the business of turning intent into deployed frontends. Framing them as partners toward a 'sites partner ecosystem' signals that OpenAI wants to sit above the hosting and page-building layer rather than only feed code into it.
For teams building on any of these platforms, the practical question is where the generation boundary lands: whether Codex produces the frontend and hands off deployment, or whether it increasingly owns both. The announcement doesn't resolve that, but naming these companies as collaborators while shipping a competing capability is the tension worth watching.
The product design plugin works from live UI, not just specs
The product design plugin is built for turning early ideas into prototypes teams can review. Two capabilities stand out for frontend work: prototyping from a live URL, and making static screenshots interactive. Both treat existing interface state as input rather than starting from a written brief, and work carries forward into Figma and Canva.
This matters because it lets Codex meet designers where their artifacts already are—an existing page or a flat screenshot—instead of requiring everything to be described in prose first. It's the same 'work with the tools your team already uses' logic OpenAI applies across the six new plugins, which together bundle 62 apps and 110 skills.
Annotations bring pointed, direct-manipulation editing to frontend artifacts
Annotations, previously used by developers to refine code, Markdown, and generated websites, now extend to documents, spreadsheets, and slides. The mechanic is selection-based: point to a specific element and tell Codex what to change. The source's own example is telling for frontend teams—select the navigation bar in a site and ask Codex to update the font, or mark a chart and ask for a clearer label.
This is closer to direct manipulation than to prompting. Rather than re-describing an entire page to change one component, you scope the edit to a selected region and preserve everything else. OpenAI frames annotations as most useful after the first draft, 'when the work needs judgment, feedback, and iteration'—which is exactly where frontend work usually lives.
What building hosted frontends inside Codex asks of teams
The specific implication here is governance, not novelty. Sites are hosted and shared by URL across a workspace, and admins must enable them in Enterprise settings while also controlling underlying app permissions for the role-specific plugins. Once Codex can generate interactive pages that read from Snowflake, Salesforce, or a financial model and then keep them current, those pages become live surfaces exposing internal data—not throwaway drafts.
With non-developers already about 20% of Codex users and growing more than 3x as fast as developers, the people creating these frontends increasingly won't be engineers. The immediate work for applied teams is deciding what a Codex-generated, auto-updating site is allowed to display, who can share it, and how its data connections are scoped—before a URL circulates that no one on the platform side reviewed.
Find this story relevant to you?
Contact us to find a unique solution
Need an AI engineering partner that can actually build?
We help businesses integrate AI, build AI-powered products, automate high-value workflows, and modernize the software systems behind them.
Related reading
More analysis around product delivery, operational AI, and the systems work that makes deployment hold up in reality.