Artifact

The word 'artifact' predates AI: in software and archaeology it means something created as a result of a process. In AI builder discourse it has taken on a specific meaning: the discrete, reusable output of a model interaction. Not the conversation itself, but the thing that came out of it. A draft contract, a working React component, a data analysis script, a logo in SVG. Something you can hand to someone or deploy.

Claude's Artifacts feature made this concept concrete and visual. When Claude produces substantial self-contained content, it pops it into a dedicated side panel where you can preview, iterate, download, and share it directly. The artifact is rendered and live: an interactive app is clickable, a chart is readable, code runs. OpenAI's Canvas is a similar idea with a different interaction model, focused more on collaborative editing.

The broader term matters beyond Claude: when builders talk about 'prompt-to-asset' workflows, CI/CD pipelines that generate artifacts, or agentic tasks that produce output artifacts, they're using the same concept. The model's job is to produce something useful, not just to answer. That shift from chatbot to artifact-generator is a big part of what makes AI feel like a tool rather than a toy.