Documentation Index
Fetch the complete documentation index at: https://university.gumloop.com/llms.txt
Use this file to discover all available pages before exploring further.
The AI industry is horrendous at naming things. MCP, o4, Mini. The most used AI tool is literally called ChatGPT. Another poor naming decision is artifacts, what we call files generated by an agent. In normal speak, we’d call that a spreadsheet, a presentation, a doc, a dashboard. Frankly, we’d call it a file. I promise that’s it for my rant. Let’s talk about artifacts.
So I’ve got a sales agent here, and I’ve just asked it for an update on our sales pipeline. And it gave it back to me directly in the chat. So let’s go ahead and create an artifact with something like, “Generate a dashboard of our pipeline.” And it’s going to go ahead and write a little bit of code and generate me a dashboard, an artifact.
Now, the reason agents can build dashboards is that fundamentally, a dashboard is a website. So is a spreadsheet. Anything in the cloud is a website. And you can generate all of these different file types with Gumloop agents.
Now, if I go back to my artifact, each artifact is hosted in the cloud. So it has a URL. By default, it can be viewed by anyone who has access to the agent. And I can expand who has access by sharing it with that person specifically, sharing it with my organization, or even making it public so anyone has access to it. This works similarly to Google Docs, and it’s even got versioning built in. So if I ask my agent to change anything, like let’s go ahead and make this dashboard light instead of dark, it’s going to go ahead and create a new version of the artifact.
Now, just from this sales agent, I can have a spreadsheet of leads, a PDF executive summary I can share with the team, all based on the real data in my CRM.
Part two, let’s talk about AI slop. We’ve actually named this one correctly. This is where you let the AI generate a report, or frankly any content, without guidance. And well, what’s going to happen is that it’s not really going to look on brand. It’s got this purple hue. It just generally looks like, well, AI generated. And not only that, but every time you ask for a new report outside of this chat, it’s going to look a little different because agents have no memory of what they created before.
Now, with skills, we can fix both of these problems. Generate something relatively on brand, and have the AI consistently reproduce an artifact in a new chat. So for the styling part, let’s go ahead and give access to Firecrawl. That will allow our agent to scrape our site and understand our styling. And let’s say something like, “Grab the design from Gumloop and redesign this dashboard, then write a skill so when I ask you for a pipeline report, you recreate this dashboard every time.”
Now, we’ve got our new artifact looking much nicer. And in the skill, if I open that up, we have the exact code that the agent needs to generate it again. So now when we ask it to generate a pipeline report, it’ll know what to do because it’s in the skill.
Don’t let the name scare you, or stop you even, from getting more from your agents using artifacts. Go out there, create a spreadsheet, a doc, a report, whatever you need for your work.
Artifacts
Artifacts are files your agents generate and host in the cloud. Dashboards, spreadsheets, reports, presentations. Anything you’d normally build by hand, your agent can create for you.
When you ask an agent a question, it usually replies in the chat. That works for quick answers, but sometimes you need something more structured. A dashboard, a spreadsheet, a PDF. That’s what artifacts are for. Your agent writes the code, generates the file, and hosts it at a shareable URL.
What agents can generate
Anything that lives in a browser is fair game. Dashboards, spreadsheets, slide decks, reports. Under the hood, these are all just web pages, which means your agent can build them from scratch using the data it already has access to.
A sales agent connected to your CRM could generate a pipeline dashboard, a lead spreadsheet, and a PDF summary, all from one conversation.
Sharing and versioning
Every artifact gets its own URL. By default, anyone with access to the agent can view it. You can also share it with specific people, your whole organization, or make it fully public. This works the same way you’d share a Google Doc.
Artifacts also support versioning. If you ask the agent to change something (“make this light mode instead of dark”), it creates a new version rather than overwriting the original.
Fixing the “AI slop” problem
Without guidance, generated content tends to look generic. Off-brand colors, inconsistent layouts, the kind of output that screams “AI made this.” And because agents don’t remember past conversations, the next report will look different again.
Skills solve both problems. Connect a scraping tool like Firecrawl so the agent can study your brand’s design. Then ask it to restyle the artifact and save the approach as a skill. From that point on, every time you request that type of report, the agent reproduces it with the same styling, in any new chat.
What to remember
Artifacts turn your agents into content producers. Instead of copying data out of a chat and building a spreadsheet yourself, let the agent generate and host it. Pair artifacts with skills so the output stays consistent and on brand every time.
