Anthropic just dropped something that might actually change how engineering teams communicate. On June 18, they launched artifacts in Claude Code — a feature that takes the messy, iterative work happening inside an AI coding session and turns it into a living, breathing web page that anyone on your team can open, watch update in real time, and actually understand.

If you’ve ever sat through a standup where someone tries to explain what the AI agent found while debugging a 3 am incident, or watched a PR review devolve into “wait, which file was that again?”, this is aimed squarely at you.

Artifacts are shareable, interactive web pages that Claude Code builds from the full context of your session — your codebase, your monitoring tools, your conversation history, the works. You ask for a page, and it appears at a private URL. As the session continues, the page updates in place. Same link, new version, full history. Only people in your organization can see it. It’s in beta now for Team and Enterprise plans, accessible from both the CLI and the desktop app.

Why this feels different

We’ve had AI coding assistants for a while now. They write code, they explain things, they even help debug. But the output has always been trapped in the chat window — scroll back, copy-paste, lose context, repeat. Artifacts break that container.

Think about what a typical Claude Code session actually looks like. You’re investigating an incident. The agent pulls logs, traces through your codebase, checks the monitoring dashboard you connected, identifies a suspect commit, and reasons through the root cause. Normally, you’d get a wall of text in the terminal. With artifacts, you get a page: a timeline, the relevant code snippets, an error-rate chart, the suspect commits — all clickable, all sourced from the actual work.

Then you share the link. Your teammate opens it. They see exactly what you see. When the agent publishes an update five minutes later because it found something new, the page refreshes. No “let me walk you through it.” No stale screenshots in Slack. The artifact is the status update.

Built on context, not infrastructure

This is the part that’s genuinely clever. Anthropic didn’t build a new dashboarding tool or a BI platform. They realized the AI already has everything it needs — the codebase, the connectors (monitoring, logging, infra), the conversation. The artifact is just a rendering of that context.

You don’t wire up data sources. You don’t stand up infrastructure. You say, “build me an incident page with the timeline, the failing test, the error spike from Datadog, and the root cause” and it happens. The agent reaches into what it already knows and builds the page.

The same logic applies across roles. A privacy engineer asks for a data-flow map of where personal data moves through the codebase. A FinOps person asks for cloud resources and cost drivers mapped from Terraform. A staff engineer wants a real import-graph visualization of how the payments service actually fits together — not the architecture diagram from two years ago, but the real thing, right now.

Live pages that don’t pretend to be apps

There’s an important constraint here that’s worth understanding. Artifacts are standalone HTML pages. No external network requests. No APIs. No WebSockets. No form submissions. All CSS and JavaScript inlined. Images as data URIs. Hard size limit: 16KB rendered.

This isn’t an accident. Anthropic is explicit: artifacts capture work, they don’t become software. They replace whiteboard diagrams, manual bug walkthroughs, status reports, and those “quick Loom video” recordings that nobody watches. They’re visual, self-updating summaries that live inside your org’s trust boundary and don’t leak data.

If you need a real app with a backend, build a real app. If you need the team to understand what happened during the incident, what the PR actually changes, or why the cost spiked — that’s what artifacts are for.

Private by default, shareable on purpose

Every artifact starts private to its author. You share it when you’re ready, and only with authenticated members of your organization. Admins get an org-level toggle, role-based scoping, retention policies, and compliance API visibility. It’s enterprise-grade without being enterprise-bureaucratic.

Version history is built in. Every publish creates a new version at the same URL. You can restore old versions. A gallery lets you browse and manage everything you’ve made. It feels like a modern doc tool, not a dev tool bolted onto a chat interface.

The announcement lists examples by role, and they’re surprisingly practical:

• Legal and open source: License audit of every dependency, flagging copyleft, straight from the repo.
• Privacy: Data-flow map of personal data across the codebase for the privacy review.
• Security: Findings linked to exact lines so the fix is unambiguous.
• FinOps: Cloud resources and cost drivers from Terraform, grouped by service.
• Software engineers: PR walkthroughs reviewers can actually follow — diff, reasoning, what was tested.
• Designers and frontend engineers: Multiple UX variations built from your real component library, so the one you pick is shippable.
• Staff engineers and architects: Service maps drawn from the real import graph instead of a whiteboard.
• SRE and on-call: Incident pages that grow as you investigate and become the postmortem.
• Engineering managers: What actually shipped this week, built from merged PRs, grouped by project.

These aren’t hypothetical. They’re the kinds of artifacts Anthropic’s own teams have been using during internal testing. The debugging workflow — incident investigation before standup, artifact published, updated twice more as the investigation progresses, team already looking at the same view by standup — that came from real usage.

Artifacts first showed up in the consumer Claude web interface back in summer 2024. People built games, dashboards, interactive visualizations — the “look what I made in 30 seconds” energy was real. But that was a playground. Moving artifacts into Claude Code, with access to the actual codebase and connectors and organizational context, changes the category entirely.

It also sits alongside other recent Claude Code releases: routines (reusable prompt workflows), centrally managed MCP connector auth, workload identity federation for secure platform access. The pattern is clear — Anthropic is building Claude Code into a proper enterprise development platform, not just a coding assistant.

If you’re on a Team or Enterprise plan, you have access now. Update Claude Code, open the CLI or desktop app, and ask for something visual. “Build an artifact walking through this PR.” “Map our cloud costs from Terraform.” “Give me five UX variations of this signup form using our component library.”

The agent builds the page, gives you a link. Open it in any browser. Share from the header. Updates publish to the same URL automatically.