Figma, Excalidraw, Illustrator — and where an agent fits
If you are trying to work out whether an AI agent can drive your design tool, the short version is: mostly, no — and where it can, it is bolted on.
Here is the honest landscape, including the parts that don’t flatter us.
Figma
The best collaborative design tool there is, and it isn’t close. Real multiplayer, components, variants, auto-layout, a plugin ecosystem, and a team of people who have thought harder about design workflow than we have.
Its plugin API is genuine and powerful, and an agent can drive it — but it lives inside a plugin sandbox in a logged-in app, which means a real account, a real file, real permissions, and a plugin someone installed. That is a sensible architecture for a company’s design system. It is a lot of ceremony for “draw me a box and an arrow.”
Use Figma if you have a team, a design system, and files that matter. Not us — we have no accounts, so we have no teams, no permissions, and no private files.
Excalidraw
Wonderful, fast, and the closest thing to us in spirit: open a tab, draw, share a link. If you want a hand-drawn-looking diagram in twenty seconds, use Excalidraw.
The difference is what the drawing is. Excalidraw is a whiteboard — deliberately loose, sketchy, and not trying to be a precision vector editor. There is no bezier pen, no boolean shape building, no variable stroke width, no artboards.
Use Excalidraw if you want a sketch. Use us if you want a vector document — paths you’ll refine, a logo you’ll carve out of overlapping shapes, something you’ll export as clean SVG.
Illustrator
The real thing, and thirty-five years deep. Anything we have, it has more of. It also costs money, needs installing, and takes a while to open.
Its scripting layer is old and powerful and mostly used by people who have already committed their career to it.
Use Illustrator if you are a professional illustrator. That is not a joke and not a hedge.
Where we’re different
Not “better” — different, in one specific way: the agent is a first-class seat, not a plugin.
There is exactly one set of verbs in justdraw.fyi, and everything sends them. The
keyboard sends group.create. The toolbar sends group.create. The jd CLI
sends group.create. A language model sends group.create. There is one
implementation of what grouping means, and the machine interface cannot drift
from the product, because it is the product.
Which produces a few things that are hard to get any other way:
- Paste a URL at a model and it can work. No key, no account, no plugin, no
install. It fetches
/p/<uuid>.md— a briefing generated from that live project — and starts sending commands. - The agent’s edits are real edits. They land in your undo history, appear on your screen as they happen, and autosave like yours. You can undo the machine.
- You can watch. Keep the tab open and the drawing assembles in front of you.
- It draws objects, not pixels. If the circle is four pixels too far left, you drag it four pixels right, because there is a circle.
The costs, plainly
- No login means no privacy model. Whoever has the link can edit the drawing. For a box and an arrow that’s right. For your unannounced roadmap it isn’t.
- A drawing command needs the tab open. The browser is the engine; with no tab you get a 409. That is the price of having one implementation instead of two.
- No multiplayer cursors, no components, no version branching. Figma has spent a decade on those. We have not.
So
If an agent drawing real, editable vector shapes is the thing you need, there is not much else to try, and you can find out in about ten seconds.
Make a drawing, paste the link at a model, and see. The protocol is on the agent protocol page.