The models arrived. The workflow didn't.

Almost every creative person you know uses generative AI. Almost none of it is in production. That gap is the most interesting problem in creative software right now, and most of the industry is answering the wrong half of it.

The answer everyone reached for was more models, bundled together, behind one subscription. It's a reasonable business. It just doesn't touch the actual problem. A model produces one output when you ask it nicely. Production needs something else entirely: the same output again, on brand, in every format and language, on a deadline, approvable, and deliverable. Hundreds of coordinated things — not one good image.

What sits between a model and production is workflow. Today that layer is code. Most creative teams don't have a programmer, there's no training that teaches this well, and the tools ship without the harness that would make it unnecessary. So the wiring gets done by hand, once, per job — or it doesn't get done, and AI stays a pile of disconnected toys next to the actual pipeline.

A prompt is a wish. A graph is a decision.

We think the unit of creative work should be a thing you can re-run, inspect, hand to someone else and change your mind about. Not a lucky result you can't reproduce and can't explain.

That's what a procedural graph is for. It's a record of decisions — this drives that, this varies, this never does — written down in a form that executes. You can open someone else's and understand it. You can change something near the beginning and watch the end update. You can hand it to a colleague and know they can't accidentally break the parts that matter.

AI is a node. Not a replacement for knowing what you're doing.

Generation belongs in that graph, wired to the thing it feeds — a material on a product, a treatment on a layout, a variation across a set. Not in a panel off to the side that returns something different every time you press it. When it's a node, it becomes art-directable and repeatable. When it's a panel, it stays a slot machine.

The wrong turn: making pictures of things

Here's the deeper thing, and it predates AI by twenty years.

Creative software got extremely good at producing material — images, video, files you hand to someone else to place somewhere. That made sense when the destination was print, or broadcast, or a billboard. The output genuinely was a file, and someone else genuinely did put it somewhere.

But most creative work now ends up somewhere interactive. A page. A configurator. A screen in a space. A thing on a phone that responds to being touched. And our tools still hand us a flat picture of that thing, which a developer then rebuilds for real, in a different language, in a different building, on a different timeline.

We've all just accepted this. A designer makes a beautiful mock of an experience, and then the experience gets made again by somebody else, and the two drift apart forever. That is a strange way to work, and it's only normal because it's old.

The output should be the thing

So nooodles has two ways out of the same graph, and we're careful about the difference.

A workflow makes material — every ratio, every locale, every cut. Your team takes those files and places them. That job is real, it's most of what teams need this week, and we do it in the shape you'd expect.

A scene isn't material at all. It's the thing itself: the page someone browses, the configurator they play with, the installation they walk up to. It ships as a link. Your audience opens it and uses it.

The distinction isn't how long each one lasts — it's who opens it. And the reason it matters is that once both come out of one graph, they can't drift apart. Change the product, and the campaign and the destination are both right, because they were never two projects.

Build it once, and get out of the way

The other thing we believe is that the person who can build this shouldn't spend their life making versions of it.

In most teams, every variation routes back through the one person who can open the file. They become a queue. The work is beneath them, they know it, and everyone waits — so most of the variations simply never get made. That's not a tooling inconvenience. It's the reason the content most brands actually need doesn't exist.

So the graph carries its own interface. The author decides what varies — a colourway, a headline, a market — and locks everything else. Everyone else gets a form. Off-brand isn't a thing they're discouraged from doing; it's a thing the tool can't produce. One person builds it once, and then steps out of the loop.

Why the browser

Because that's where the work goes, and because a link is the best distribution mechanism ever built. No install, no version, no platform, no account. You send it and it opens.

And because if the destination is a web page, the tool should be able to make a web page — real elements, real type, real semantics — rather than painting a picture of one onto a canvas. That's why the document is a first-class part of the graph here, alongside geometry and materials and signals. A project that never touches 3D ships with no 3D in it at all.

Some things we're not doing

We're not training models. That's a capital race with brutal economics, and it obsoletes itself every few months. We orchestrate what exists, we're model-agnostic on purpose, and we ride the cost curve down instead of paying to push it.

We're not pretending the work isn't yours. Some client material legally can't touch a public AI service. nooodles runs where you choose — hosted when you want zero setup, on your own hardware when the contract demands it. And because the workflow is an explicit graph a human authored, the result carries documented creative direction, which is the strongest authorship position AI-assisted work can hold.

We're not going to tell you it's finished. This is early. Things move, things break, and some of what's described across this site ships with the beta rather than today. We'd rather say that plainly than have you find out.

What we're actually after

A creative person should be able to build a real, living, interactive thing — with AI in it where AI helps — and put it in front of people the same afternoon. Without a programmer. Without a hand-off. Without it being a picture of the thing.

That's it. That's the whole idea.

— Jacques Parys, Toronto