Claude Artifacts: stop chatting, start shipping

6 min read  ·  Artifacts explainer  ·  Last updated 13 May 2026

Part of Learn Claude Code: The Complete Operator's Guide. For the operator's overview of Skills, Connectors, Cowork, and Artifacts, start there.

You ask Claude for a one-page proposal. It writes back six paragraphs in the chat. You copy them, paste into Google Docs, reformat, struggle with the heading hierarchy, give up and rewrite in your own template.

That is the chat-only workflow. It works, but every output costs a translation step.

Artifacts skip the translation. Claude builds the proposal in a side panel as a real document. You edit it in place, iterate with chat, and download or share when it is ready.

What an Artifact actually is

When you ask Claude for something substantial, the chat splits into two halves.

Left side: the chat conversation, like normal. Your prompts, Claude's replies.

Right side: the Artifact. The actual document, code, chart, or interactive thing. Editable. Downloadable. Shareable as a public link.

Claude decides when to use an Artifact. Anything more than a few paragraphs and it usually pops one open. You can also force it: "make this an artifact" works.

When to reach for Artifacts

Think about what you do today when Claude writes you a proposal. You copy the text. You paste into Google Docs. You re-format the headings. You insert your logo. You fix the spacing. Five steps before you can send anything. Then you find a typo, ask Claude for a fix, repeat the five steps.

Artifacts compress that whole loop into one step. Claude builds the doc in a side panel, you edit it in place, you share or download when ready. The five steps become zero.

The rule of thumb: if the output is something you will share, edit, or look at later, it belongs in an Artifact.

Reach for them when:

Stay in chat when:

Three things you should always make with Artifacts

1. One-page docs

The single biggest unlock for non-developers. Proposals, briefs, meeting agendas, executive summaries, project plans.

Ask Claude for the doc. Edit in the side panel. Download as a PDF or markdown. Send.

The whole loop is a quarter of the time it would take in Google Docs, and the output reads better because you are iterating directly with the model that wrote it.

2. Quick charts from data

Paste a table from your CRM, your spreadsheet, or your call log. Ask Claude to chart it.

The chart pops as an Artifact. You can ask for changes ("make it a bar chart", "highlight the top three") and it updates in place. Screenshot or share-link when you are done.

Faster than opening Excel, building a pivot, choosing a chart, formatting.

3. Small interactive tools

This is the unexpected one. An ROI calculator. A pricing comparison page. A quiz for your team. A landing page mock-up.

Claude can build a small HTML app as an Artifact, complete with working logic. You can use it immediately in the side panel, or grab the code and ship it somewhere.

We have built more "throwaway" small tools as Artifacts in the last year than the previous five combined. Most never make it to production. Some do.

How to share an Artifact

Open the Artifact. Hit the share button. You get a public link.

That link works for anyone, whether or not they have a Claude account. The Artifact renders as a live, interactive view of whatever you built. Code becomes a working preview. Documents become reading view.

This is how a small calculator goes from "I made this in 10 minutes" to "the team uses this every Friday" without ever leaving Claude.

Live Artifacts: the Cowork upgrade most people miss

Anthropic recently shipped a stronger version called Live Artifacts, specifically for Cowork. The official definition: "A persistent, interactive HTML page that Claude creates for you in Cowork. A tracker, a dashboard, a comparison tool, a reference, shaped around your specific work."

Three things make a Live Artifact different from a regular chat Artifact:

1. They persist

Live Artifacts get their own tab in your Cowork sidebar. You do not have to find the chat they came from. Open them like any other document, anytime.

2. They refresh themselves

When you open a Live Artifact, it can pull live from your connected apps and local files. The view reflects today, not the day Claude built it. A dashboard built on Monday shows Tuesday's data on Tuesday.

3. Version history

Every update saves a version. You can review how the artifact has evolved over time and restore an earlier version if a change broke something.

Five Live Artifacts Anthropic ships as examples

Notice the pattern. Live Artifacts are most powerful when they read live data from things you already use. The Connector layer (Notion, Gmail, Linear, Drive, your CRM) is what feeds them. Connectors 101 covers the wiring.

How to create a Live Artifact

Two ways:

Each Live Artifact gets a manual refresh button in its header if the auto-refresh is not pulling fast enough for your taste.

Safety: read this before you build one

Anthropic flags one important caveat: "Live Artifacts use your connectors without asking on every refresh."

That means: if your Live Artifact has a connector with write or delete permissions attached, it can fire that connector every time the artifact refreshes. Three safety habits worth building from day one:

Also: Live Artifacts are not shareable yet. Sharing is on Anthropic's roadmap. For now they are for your own use, and they replace the "I built a thing, now let me reformat it as a Notion page" loop.

Artifacts plus Skills: the multiplier

Artifacts on their own are good. Artifacts with a skill attached are how Waboom ships at speed.

We have a brand-guide skill that knows our colours, our font choices, our voice rules, our footer template. Every Artifact Claude builds inherits all of that automatically.

The result: a one-page proposal Claude generates as an Artifact is already on-brand. No reformatting. No "make it match our colour scheme" prompt.

This is exactly the skill we covered in What is a Claude Skill. The combination is where the leverage compounds.

Common stumbles

Forgetting Artifacts exist. Most people default to chat-only. The simplest unlock: next time you would copy-paste Claude's output into a doc, ask "can you make this an artifact instead?"

Treating Artifacts as final. They are designed to iterate. Highlight a section in the panel, ask Claude to change it, watch it update in place. Faster than rewriting.

We run this exact iteration drill at our Claude Code course. By the second hour every attendee has a brand-styled Artifact they built themselves, paired with a skill that styles every future Artifact the same way.

Not using the share link. The share link is a real product. Treat the Artifact as something you would actually send to a client, because you can.

What to do next

Open a chat. Ask Claude for the most boring document you have to write this week. Force an Artifact ("make this an artifact"). See how the workflow changes.

Then layer a skill on top so the Artifact comes out in your voice and your format automatically.