Artifacts in Claude Code

Artifacts in Claude Code

scott cunningham

Oct 30

The thing about Anthropic regularly updating its product, Claude, is that it sneaks in fairly interesting, sometimes what feels like radical, changes. But if you’re like me, you just click the “update” button that shows up on the desktop app, and rarely think to dig into the changes. At most, I just sort of notice something aesthetic is different on the desktop app, but otherwise I don’t know what I don’t know and so only learn anything substantial has changed if something random happens. And I think something substantial has changed, but maybe you already knew this, in which case ignore this post.

The thing I want to talk about are “artifacts”. I’ve been making artifacts on Claude for a very long time, as I went through them the other day, and it looks like the first time I made one was in August 2024 for one of my classes. You can find Artifacts top left of the desktop app (go ahead and upgrade if you haven’t in a while).

What exactly is an “artifact”? Listen to how Anthropic defines them here.

An artifact is anything Claude makes for you that you’d put in front of someone: a design, a deck, a document, a dashboard, or a small interactive tool. It opens beside your conversation, and you can edit it, come back to it, and share it with others. Ask for one in any conversation with Claude, including in Claude Code, or start from a template in the “Artifacts” tab.

That “anything” is the key to understanding them. They are quite literally anything you can conjure up the imagination to create, and I think when I realized that, that was what opened them up for me.

Most of the artifacts I have made appear to be decks that I did not mean to make. I prefer to make decks in beamer on my local machine — force of habit —although recently I’ve begun making them in html. And probably when I switched to making my decks in html, I think I accidentally started making more artifacts without realizing it. They would show up regularly with my instructions, almost like they were being opened in an internal browser that is Claude proprietary, which I think I just assumed was the localhost:8080 happening displaying my own local html deck. I would simply close it, not really thinking about it.

Until last weekend.

Last weekend, my coauthor came into town, and we spent 72 hours finishing up this paper we’ve been working on for years. I don’t think we would’ve finished it in 72 days, 72 years maybe, otherwise. It taught me again that being physically together to complete papers almost has no price tag if the goal is to finish the paper. But put that aside.

I wanted to share with him an html deck walking us through the paper, since it had been a while. And I thought “well wouldn’t it be interesting if we could both vote on each figure and table as to whether we felt we could keep it or get rid of it?” So then I thought, “well, people do this on their websites, so I bet Claude can make it”. And therefore I requested that — a button for me and my coauthor to vote on after we each reviewed each figure and table.

Well, that was when I noticed it wasn’t an html at all; it was an internal Claude artifact with a share button. Here is a snapshot of the share button. You put the person’s name in, they get a link, and then the artifact becomes live for us both. It becomes live in the sense that the other person can interact with it, and so can I.

When I realized that, I then realized that just like Claude can do anything I can think up, I could do anything with this that I could think up. So, I had an idea. Let’s lay out the figures and tables, with a blind voting mechanism, as well as a place to leave comments. And then let me use the /loop command to pull the material from the artifact on a regular basis and do whatever it said. Something like this for instance:

/loop [5 min] [pull the artifact, do the things in the edits, push it back, recompile]

And when I started piecing that together, I wondered to myself — well couldn’t I use the /loop command in combination with boxes for text to allow us each to do things on my machine? For instance, if we felt like there was something strange about a figure, couldn’t we leave comments with clear instructions of what we wanted to see, and then tell Claude to maybe insert it somewhere in the artifact for examination? Could I make each page of the artifact moveable? Could I be able to drag objects around? Could I do something really weird — could I have Claude make a video with audio that explained key portions of the deck, such as the policy change, and paste that in? Could it do that while overlaid with animation?

And the answer is yes. The answer is almost always yes to any question you ask Claude to do. You are like Green Lantern with his magic space ring — Claude is only limited by your imagination, not limited by its ability.

So now I have been shifting a bit away from skills to more interactive things inside the artifacts. Or let me say it differently — artifacts seem to be a way for me to do some of the things I thought I was only going to be able to do with a skill. Like what? Like checking references.

My current goal is to go through each references one at a time, checking both how the cite was cited, whether it makes as much sense now as it did then, as well as to check the accuracy of the cite in the .bib itself. I wanted a vote button, I wanted the ability to read the context on the paper, a link to it online, so that I could just trace one to one. I’m a visual person, I need the actual physical experiences of checking things, but as I’ve talked about a lot, I’ve been trying to find workflows that are less automated and more augmenting, but amplified, and figuring that out — I’ve sensed I personally needed checklists and dashboards, but none of them have been quite what I’ve wanted.

I’m not saying that artifacts are the key, but for final checks, I’m using them more. For making more shared experiences with coauthors, such as voting, I’m doing it. I’ve been toying more lately, because of a different project, if I might be able to even train a language model for classification using something like artifacts which are shared with my coauthors, which when we’re done is shared back to Claude and can move into the classification stage.

So check it out. Tinker with it more than you were. See if it’s helpful. I get the sense that Anthropic thinks that they are given they keep investing more in them. I am starting to wonder if they want users to spend more time in the desktop app, too, and less time in your favorite IDE. I wonder if the equilibrium is a very functional IDE experience within the desktop app. Maybe they’re trying to build more of a moat around the product by building more user experiences in that are Cursor or VS Code domains. Whatever it is, though, I do think this could be where things are going for deck making, perhaps putting some dings in the armor of Apple Keynote and Microsoft Powerpoint.

© 2026 scott cunningham
n/a