Anthropic has added a new feature called "Artifacts" to its AI coding tool, Claude Code. It generates a live HTML page on the spot from a Claude Code session and lets you share it with teammates. The feature is available to users on the Claude Team and Enterprise plans, and it targets use cases such as dashboards, diagrams, and progress reviews — turning the contents of code into something even non-engineers can understand.
What Artifacts Is
Artifacts takes the work Claude Code is doing and turns it directly into an interactive web page. Drawing on the user's local code (repository), connected monitoring tools, and the running conversation, Claude Code assembles a dedicated page. Engineers no longer need to set up an external data platform or stand up temporary servers. The AI builds the interface (UI) from materials that already exist.
Each generated page gets its own URL, which can be sent to teammates. Recipients can watch Claude Code make progress — autonomously or under guidance — right there on the same page.
The Artifacts name itself is not new. Anthropic first introduced a feature of the same name in its consumer chat interface in the summer of 2024. That mechanism spread as a tool for publishing code snippets and games to the web; this time, it is built directly into the developer-facing Claude Code, which is operated from the command line, and into the desktop app.
A "Live" Page, Not a Static Export
The defining trait of Artifacts is that it is not a one-time export. As the AI works through a session, the open page updates in place at the same URL. Charts and text are swapped out in real time, and anyone watching sees the latest state.
Because a version history is recorded with every update, you can roll back to an earlier state or check how far the AI has progressed. The page can be viewed on a PC or a smartphone. The goal is to replace meeting whiteboards, manual bug walkthroughs, and routine status reports with secure visual tools that update automatically.
How It Differs from OpenAI's "Codex Sites"
This announcement comes about two weeks after OpenAI added a similar enterprise hosting feature, "Sites," to its own Codex. The two companies have been rolling out comparable developer work surfaces one after another. Their design philosophies, however, differ clearly.
OpenAI's Sites is built as a foundation for running full-fledged web applications. It includes mechanisms for storing data in a database and for file storage, and it supports external login integration and fine-grained access controls. It is positioned as a persistent runtime environment that can replace internal business tools outright.
Claude Code's Artifacts, by contrast, deliberately has no backend server functionality. In its official explanation, Anthropic states plainly that "an artifact is a capture of work, not an application." Each page is a single, self-contained HTML file, with a cap on rendered size (about 16 MiB).
Furthermore, to protect organizational security, generated pages are wrapped in a strict policy (a Content Security Policy). External communication is blocked, so the page cannot load external scripts, fonts, or stylesheets. All CSS and JavaScript must be inlined, and images are embedded directly as data. The page cannot store form input or call an external API at view time.
This restriction is not a weakness but Anthropic's deliberate stance. Whereas OpenAI aims for persistent software portals, Anthropic keeps Claude Code anchored to "ephemeral, highly secure technical work." The idea is a self-updating visual tool from which live data never leaks outside the corporate boundary.
Security and Access Management
Because these features touch a company's proprietary data and live code, licensing and access management matter. Both Artifacts and Codex's Sites adopt closed, provider-managed models; client companies cannot freely copy the internals or run them on their own servers.
Anthropic makes each artifact private to its author by default and designs it so that it cannot be published to the open internet. Even when a link is shared, only authenticated members of the same organization can view it. Administrators can govern things through org-level toggles, role-based scoping, and retention policies, and they can oversee the whole picture through a dedicated management interface (a compliance API).
OpenAI likewise limits Sites to ChatGPT's Business and Enterprise tiers and uses a system in which administrators enable the feature by role. It also offers more granular control over access scope when sharing, plus a dedicated panel for handling sensitive information without writing it directly into source code.
How It Is Used in Development
Boris Cherny, the lead and creator of Claude Code, said in a post on X to coincide with the announcement that he uses Artifacts in every part of his daily work. He cited visual explanations of tricky code, system diagrams, quick comparisons of several options, data analyses, and dashboards shared with the team, calling it a feature that changes the way he works.
The example Anthropic published is concrete, too. In one scenario, an engineer asks Claude Code to investigate whether user drop-offs have risen since a recent release. Within seconds, the AI reads the database, assembles a dashboard showing the drop-off funnel, and pinpoints that "Pro accounts stall at the export sheet." It then proposes UI improvements, updates the charts as the code is refactored, and generates a secure link that a manager can open right away on a smartphone.
It is an attempt to turn the terminal (the text-based screen used for operations) into a live canvas for collaboration. You can see the direction here: the most valuable output of an AI coding assistant is not just the code itself, but making the context and the reasons behind decisions instantly shareable.
Summary
Claude Code's Artifacts is a new feature that lets you share the contents of your work with the AI as a live HTML page that updates automatically at the same URL. Its defining characteristic is that it deliberately forgoes backend server functionality and imposes restrictions that block external communication, narrowing its focus to highly confidential technical work. This is a contrast to OpenAI's Codex Sites, which pushes toward persistent web applications, and the difference in the two companies' philosophies is on clear display. It is a modest addition compared with the launch of a new AI model, but as a foundation for quickly conveying development results to an entire team, its practical impact looks far from small.
