How AI actually edits your documents: living artifacts explained

AI edits your documents by generating them as structured, versioned objects — called artifacts — that live outside the chat and can be reopened and revised, instead of pasting a wall of text into a message that you then copy out by hand. That distinction is the whole trick: an artifact is a real file with a real edit history, not a chat bubble that happens to be long.

What's actually wrong with pasting AI output into chat?

Nothing generates in a chat message. Ask an assistant to draft a report and it hands you a block of markdown in a scroll of other conversation. Want to change the third paragraph? You either re-paste the whole thing with your edit described in English and hope the model reproduces everything else correctly, or you copy it out to another app and lose the thread entirely. There's no real document underneath — just text that looks like one.

What makes something a "living" artifact instead of a text block?

Two properties, together. First, it's structured content, not a string: a spreadsheet artifact is real cells with formulas that recalculate, a document artifact has actual headings and tables, a slide deck has actual slides — not markdown pretending to be one. Second, it's versioned: every edit creates a new version rather than overwriting the last one, so the artifact carries its full history and you can always see how it evolved or step back to an earlier draft.

one-chat.app — it builds things
Turn that into a budget I can play with.
Done — live budget on the right. Drag any number.
Japan · Aprillive
Flights$1,840
Hotels$1,310
Food + fun$900
Total $4,050 · under budget 

How does an AI edit an existing document without breaking it?

Because the artifact is structured rather than raw text, the assistant has focused editing tools per type — it can add a row to a spreadsheet, restyle a heading in a document, or update one slide in a deck without regenerating the whole file from scratch and risking it drifting from what you actually wanted kept. In 1Chat, this works across the formats people actually use day to day: spreadsheets (one CSV per tab, real =SUM(...) formulas, downloadable as a styled .xlsx), documents (headings, lists, tables, bold — downloadable as a real .docx), slide decks (downloadable as .pptx), plus code, Jupyter notebooks, mermaid diagrams that render live, and plain markdown. Ask for a change and the model edits the artifact in place; the new version replaces nothing, it just becomes the current one.

Can you edit a file you already have, or only ones the AI generated?

You can bring your own. Drag a file into the library — or use the Upload button — and it's parsed into the same editable structure the AI would have generated itself: a .docx becomes an editable document artifact, an .xlsx becomes a live sheet with working formulas, a .pptx becomes an editable deck. A .pdf becomes searchable knowledge, with the original kept for download, since a PDF isn't really an editable format so much as a reference. The type is auto-detected, and you can override it if it guesses wrong. From that point forward it behaves exactly like anything the assistant generated — reopen it, ask for changes, watch a new version appear.

Why does this matter beyond convenience?

Because it changes what "editing with AI" means. Pasting text into a chat is a one-shot generation you manually maintain afterward. A living artifact is closer to how you'd work with a colleague on a real file: it persists, it has a history, and either of you can pick it back up later and keep going. You can also share an artifact — or a whole folder of them — with an agent, granting read or read-write access, so a specialist you've built can keep editing your documents as part of its own work without you shuttling files back and forth. The document stops being something the AI produced once and becomes something you and the assistant both keep working on.