Contents

91 / 153

The Stack: Products, Workflows, and the Whole Board

You do not use a model, you use a workflow

Chapter 90

2 min read

Reviewed v78 · August 2026

Ask a working creator in 2026 which model they use and you will often get a pause, because the honest answer is several, behind an interface that hides them. The model stopped being the thing you touch.

The interface moved up a layer, to the aggregators and canvases. On Krea, Higgsfield, Flora, or a ComfyUI graph you assemble a result from many models and steps: generate here, upscale there, swap a face, extend a shot, add motion, and the specific model behind each step is a detail, often swappable. The product's value is the workflow, the taste baked into its defaults, and the friction it removes, not any single model it happens to call. Krea, a unified multi-model canvas, passed twenty million users on exactly this proposition; Magnific, now merged into Freepik under the Magnific name, built a bootstrapped business at a couple of hundred million dollars of annual revenue by consolidating the stack rather than training a flagship.

This is why the aggregator layer is where a lot of the usage and money now sits. These products own the creator relationship, and a model lab that only ships an API is renting attention from whoever wraps it. It is also why the same handful of models appear behind dozens of apps: the app is the business, the model is a supplier.

The next turn is already visible: the workflow itself is becoming something a machine drives. ComfyUI's node graph, covered earlier, made every step explicit and composable, and agentic tools like Comfy MCP let an AI agent assemble and run those graphs on your behalf. The interface is climbing from the model, to the workflow, to the agent that operates the workflow, and each step up is a step further from the commodity and closer to the customer.

The operator lesson is direct. If you are building, the defensible product is rarely a model and often a workflow: a repeatable pipeline that reliably turns an input into a finished, on-brand output, wrapped so a non-expert can drive it. That is the layer that owns the user.

Check your understanding

pass: 5 of 7

Answer at least 5 of 7 correctly to unlock the next chapter.

  1. 1. In 2026, what is the honest answer when you ask a working creator which model they use?

  2. 2. Where has the interface moved, according to the chapter?

  3. 3. What is said to be the actual source of a product's value in this layered world?

  4. 4. How many users did Krea pass on the proposition of being a unified multi-model canvas?

  5. 5. How did Magnific build a bootstrapped business worth a couple hundred million in annual revenue?

  6. 6. Why does a model lab that only ships an API end up 'renting attention'?

  7. 7. What is the operator lesson stated in the chapter?

The weekly briefing

Get the week's moves in your inbox.

A short, sourced digest of what actually moved across generative AI, every week. Free.

Free. One email a week, no spam, unsubscribe anytime. Prefer a reader? RSS.