How to Generate Design Slides and Docs with Claude Code Artifacts (2026)

Introduction

A design review usually starts with the same tedious step: copying decisions out of a repository, an RFC, and a chat thread into slides. Claude Code artifacts remove that copy step. You stay in the coding session, point Claude Code at the files that already describe the decision, and ask for a design review deck, a UI mockup, or a structured document. The result opens beside the conversation, stays editable, and can be shared by link or exported when the review is over. This tutorial walks through that loop, from a small RFC to a reviewed, exported deck.

Quick answer

In Claude Code, point the session at your repo files and RFC, ask for a design review deck, UI mockup, or document, revise it in chat or by direct edit, then publish a private link and export PowerPoint or PDF.

Evidence

The timing matters for planning. Anthropic announced on October 8, 2026: “Claude Docs, Slides, and Design are out of beta and available on every plan, including Free.” That sentence is why this tutorial does not assume a paid plan. The same rollout turns the three templates on by default for Enterprise plans on October 15, 2026, while an administrator can enable them earlier in Organization settings under Artifacts.

The scale is the second reason to take artifacts seriously. Anthropic said users had made more than 45 million docs, decks, and designs in Claude since September 16, 2026, when Docs and Slides launched in the merged Claude app. That figure covers the whole Claude product, not Claude Code alone, so treat it as adoption evidence for the artifact format, not as a promise about any single deck. A third date belongs on every migration checklist: the standalone Claude Design site closes on December 14, 2026, so work that still lives only there needs a move before then.

Two primary sources anchor the steps below. The Claude help article on artifacts defines what artifacts are and how they open, share, and export. The October 8 announcement on dashboards and motion states the general availability dates, the collaborative editing and sharing behavior, and the Enterprise default. A third primary source, the Claude Code overview, confirms that Claude Code runs in the terminal and works from the repository in front of you, which is why the repo-first workflow here works at all.

Prerequisites

• Claude Code installed and signed in, with a repository checked out locally. Any plan works, including Free, subject to that plan’s usage limits.
• One source document that already states the decision, such as an RFC, a design note, or a README section. Artifacts are only as accurate as this source.
• A desktop Claude app for the side panel view, or a browser for the terminal link.
• Permission to share the result. A private link is still a link, so do not put secrets, credentials, or customer data in the source files.

Step-by-step guide

Step 1, Put one decision in one source file

Artifacts go wrong when the source is scattered. Before you ask for a deck, collect the decision into a single file with a title, the options considered, the chosen option, the risks, and the numbers with their sources. The example below creates a short RFC for an auth token refresh change. It is deliberately small, because a short, exact source is easier to check than a long, vague one.

mkdir -p docs/rfc
cat > docs/rfc/auth-refresh.md << 'EOF'
# RFC: Shorten access token lifetime to 15 minutes

## Decision
Shorten the access token lifetime from 60 minutes to 15 minutes.
Refresh tokens stay valid for 7 days and rotate on each use.

## Rationale
A stolen access token is useful for at most 15 minutes after this change,
down from 60 minutes today.

## Risks
Mobile clients on slow networks may refresh more often.
Mitigation: retry a refresh once on a network error before surfacing an error.

## Rollout
Roll out to internal users first, then to 10 percent of traffic, then to all traffic.
EOF

Read the file back before you continue. If a number here is wrong, every slide built from it will be wrong too.

Step 2, Start Claude Code in the repository root

Open a terminal in the repository root and start Claude Code:

cd ~/your-repo
claude

Starting in the repository root matters, because Claude Code reads the files around it. If you start in a parent directory, point the first prompt at the exact file path. For a focused review, start a fresh session for the deck rather than reusing a long debugging session, because old context can leak into slide wording.

Step 3, Ask for a deck, a mockup, or a document in one prompt

Ask for exactly one artifact and name its audience, its length, and its source. A vague request such as “make slides about auth” invites invented detail. A bounded request gives Claude Code something to check against.

Turn docs/rfc/auth-refresh.md into a 6-slide design review deck for
backend reviewers. Slide 1 states the decision. Slides 2 to 4 cover
rationale, risks, and rollout. Slide 5 lists open questions. Slide 6
lists the rollout gates. Use only facts from that file. Put any claim
that is not in the file in a clearly marked "Needs confirmation" box.

The “Needs confirmation” box is the most useful sentence in the prompt. It forces uncertainty into the open instead of letting it hide as a confident bullet. When the deck appears, read that box first. If it is empty, the deck stayed inside the source. If it is not empty, resolve each item before the review, or delete the slide that depends on it.

For a UI mockup, describe elements and states, not a visual style you cannot verify: the login screen, the expired-token state, and the refresh-retry state, each labeled. For a document, ask for headings that mirror the source file, so a reviewer can trace every section back to the RFC.

Step 4, Open the artifact where you work

On the desktop app, the artifact opens in the side panel beside the conversation, so you can read the deck and keep talking to Claude Code in the same window. In a terminal session, Claude Code provides a link to open the artifact on the web. Open that link in the same browser profile and account you use for Claude, because an artifact opened under a different account can look empty or ask for access you do not have.

Once the artifact is open, check structure before wording. Check that the slide count matches the request, that each slide has one idea, and that the decision slide says the same thing as the RFC title. A polished deck that states a different decision from the source is worse than a rough deck that states the right one.

Step 5, Revise in conversation, then edit directly for small fixes

Artifacts support two revision paths. Use the conversation for structural changes, because Claude Code can rework several slides at once and keep wording consistent. Use direct editing for small fixes, because typing the correction yourself is faster than describing it.

Move the mobile retry risk to slide 3 and add one line: a single retry
on network error, then a visible error. Shorten slide 4 to three bullets.

After a conversational revision, re-read the whole deck, not just the changed slide. A moved risk can leave a duplicate bullet behind, and a shortened slide can drop the mitigation that made the risk acceptable. For direct edits, fix the typo or number in the artifact itself, then tell Claude Code what you changed, so the next revision does not restore the old text.

A reviewer can fix a line directly or leave a comment on a slide, while you keep asking Claude Code for larger restructures. Real-time collaborative editing means both paths can happen at once, so agree on ownership before the review: one person owns numbers and dates, and everyone else comments rather than overwriting them.

Step 6, Publish a shareable link and check what it shows

When the deck is ready for reviewers, ask Claude Code to publish it and give you the link. A published artifact updates as the session continues, which is convenient and risky: a fix reaches reviewers without a new file, but a half-finished edit can reach them too. Publish only after the “Needs confirmation” box is empty, and check the published link in a private browser window before you send it anywhere.

Check three things on the published view. First, the title and decision match the source. Second, no private note, credential, or internal URL survived from the source files. Third, the sharing scope matches your intent: a private link, an organization link, and a public link are different promises. If a reviewer reports seeing an old version, republish from the current session rather than sending a second link, because two links to two versions guarantee confusion.

Step 7, Export a frozen copy for the record

A live link is good for review, and a frozen file is good for the record. Export PowerPoint or PDF when the review closes, and use the editable Google Slides export when the next team needs to keep editing. PowerPoint and PDF downloads are described in the help article and the October 8 announcement, with Google Slides exports that stay editable after download, and Design exports to PowerPoint, PDF, and HTML. Pick by what happens next: PDF for a signed-off decision, PowerPoint for a meeting, Google Slides for a living document.

Open the exported file before you archive it. Check fonts, chart labels, and slide breaks, because an export can shift layout that looked fine in the editor. If a chart label is cut off, fix it in the artifact and export again, rather than patching the exported file, or the artifact and the record will drift apart. Store the exported file next to the source RFC, with the same date in the filename. For organizing sessions, see our Claude Code projects tutorial, and for extending sessions with external tools, see our Claude Code MCP setup guide. The broader MCP server tutorial explains the server side those tools connect to.

Troubleshooting

The artifact never appears after the prompt. Check the artifact settings first. On Enterprise plans, Docs, Slides, and Design turn on by default on October 15, 2026, and an administrator can enable them earlier in Organization settings under Artifacts. On other plans, confirm that code execution and file creation are enabled, because an artifact needs both to render. If the setting is on and nothing still appears, start a fresh session with a shorter prompt and a single artifact type.

The terminal link opens an empty or inaccessible page. You are almost certainly signed in under a different account, or you opened the link before the artifact finished rendering. Open the link in the browser profile tied to the Claude account that ran the session, and wait for the artifact to load. If the desktop app is available, open the same session there and use the side panel view instead.

The deck states numbers or dates that are not in the source. This is the most common artifact failure, and it is a prompting failure, not a rendering bug. Re-run the request with the source file named explicitly and the “Needs confirmation” box from Step 3, then delete any slide that still cites nothing. For numbers that change often, keep them in the RFC and regenerate the deck when the RFC changes, rather than editing the number only in the deck.

A reviewer sees an old version of the published link. A published artifact updates as the session continues, but a reviewer may hold a cached copy, or you may have edited a local copy instead of the published session. Ask the reviewer to reload the link, republish from the current session, and confirm the decision slide wording matches. If the review is closed, stop editing the live link and send the exported PDF instead.

The export looks different from the editor. Layout shifts happen most often with long titles, wide tables, and custom fonts. Shorten the title, split the table across two slides, and re-export, rather than fixing the exported file by hand. If the Google Slides export shifted, check it in Google Slides itself, not in a previewer.

FAQ

What is a Claude Code artifact? An artifact is an editable document, deck, or design that opens beside the Claude Code conversation, built from the files and context already in the session. You can revise it in chat or edit it directly, then share it by link or export it.

Do I need a paid plan to use Claude Design, Slides, and Docs in Claude Code? No. Since October 8, 2026, the Docs, Slides, and Design templates are available on all plans and on by default. On Enterprise plans they turn on by default on October 15, 2026, and an administrator can turn them on earlier.

How do I open an artifact from a terminal Claude Code session? Claude Code gives you a link to open the artifact on the web. On the desktop app, the same artifact opens in the side panel beside the conversation.

Can reviewers edit the deck while Claude is still working on it? Yes. Artifacts support real-time collaborative editing, so a teammate can fix a line or leave a comment while Claude continues in the same session. Agree on one owner for numbers and dates before the review, so parallel edits do not overwrite each other.

What happens to the standalone Claude Design site? The standalone site at claude.ai/design stays open until December 14, 2026, and then closes as the feature moves fully into Claude conversations. Move anything you still need out of standalone projects before that date, because their chats and comments will not carry over.

Which export should I send to a reviewer who does not use Claude? Send PowerPoint or PDF for a frozen review copy, and use the editable Google Slides export when the reviewer needs to keep editing. Check fonts and charts in the exported file before you send it, because an export can shift layout.

Conclusion

Claude Code artifacts work best when the source is small, the request is bounded, and the review loop is explicit: one RFC in, one deck out, conversational revisions for structure, direct edits for small fixes, a published link for review, and an exported file for the record. Keep the “Needs confirmation” box empty before you publish, and keep the exported file next to the source. Next, organize longer sessions with our Claude Code projects tutorial, and for live data, start from our Claude Code MCP setup guide.

Frequently asked questions