Why AI Creation Needs a Real Document
Generation is only the first move. The durable work starts when pixels become selectable, structured, versioned and safe for an agent to change.
pixels → system
The first wave of AI creative software optimized for the reveal: type a sentence, wait a few seconds, get an image. That interaction is powerful, but it describes the beginning of creative work—not the finished workflow.
A usable deliverable still needs hierarchy, type, masks, constraints, reusable components, brand rules, feedback and history. When the generated result arrives as a flattened artifact, the user pays a reconstruction tax before any of that work can continue.
An artifact is not a document
An artifact is something you can look at. A document is something you can operate on. It knows that a headline is text, that a card belongs to a component, that a color comes from a token and that a crop has a source.
That structure is what makes ordinary editing precise. It is also what makes an agent useful. Without it, the agent can describe a change or regenerate the entire picture. With it, the agent can target one node, change three properties and leave everything else alone.
- Selection needs stable node identity.
- Systems need components, variables and modes.
- Review needs history and reversible transactions.
- Automation needs a document it can read before it writes.
Humans and agents need the same state
Most AI features are implemented as a second interface beside the product. The human edits the canvas while the model receives a prompt, a screenshot and perhaps a small amount of metadata. The two collaborators are not actually looking at the same thing.
Fanta treats the document model as the shared interface. The human sees it through the native canvas and inspector. The agent sees it through structured state and typed operations. Both changes land in the same history.
Generation should become material
A generated image should enter the project like any other piece of material: placed on the canvas, attached to its prompt and model record, and available for selection, masking, composition and reuse.
This changes the role of models. Instead of replacing the editor, they supply it. The editor remains the place where intent becomes a system and where a team can understand what changed.
The document becomes the API
Once visual editing, source editing and agent operations converge on one model, the creative file stops being an opaque endpoint. It becomes an interface that other tools can safely extend.
That is the deeper Fanta thesis: the next design format is not merely readable by a new app. It is operable by people, code and agents without asking any of them to give up the structure the others need.
The winning AI design tool will not generate the most isolated artifacts. It will preserve the most intent after generation.
The takeaway