Artifact

An artifact is a concrete, reviewable output you can save and act on, such as a prep packet, a decision record, a brief, or a tagged note, rather than a passing chat reply that disappears when the conversation ends.

An artifact is a concrete, reviewable output you can save and act on, such as a prep packet, a decision record, a brief, or a tagged note. It has a fixed shape you can open, check, edit, and hand to the next step, so the value of the work survives past the session that produced it. Picture asking an assistant to summarize a long meeting recording. If the answer scrolls by in the chat, it is gone once the window closes; if you tell the assistant to write a meeting brief with attendees, decisions, and follow-ups into a saved note, you have an artifact you can reopen next week.

Builder example

When you build with AI, the difference between a demo and a tool is whether each run leaves an artifact behind. A summarization feature that only prints to the screen forces the user to copy and paste before the answer is lost. Instruct the assistant to write its result to a named destination in a fixed structure, such as a draft email or a dated record, and the same workflow becomes something a person can review, correct, and reuse instead of regenerating from scratch.

Common confusion: A chat reply and an artifact can contain the same words, yet only the artifact persists in a place you can return to. What makes an output an artifact is that it lands somewhere durable with a reviewable shape, so a raw model response stays an artifact only once you capture it into a saved, structured record.