Where AI belongs in a design workflow
A map of our workflow: what we generate, what we judge, and where the last ten percent lives.
- Guide
- June 2026
- 6 min read
We use AI for breadth and speed and reserve human judgment for the decisions that carry risk. Generate options and drafts with AI, but keep intent, selection, and the final ten percent human, because those are the stages that decide whether the work feels authored.
The argument about AI in design is usually framed as all or nothing: either it replaces the designer or it is a toy. Both are wrong, and both miss the only question that matters in practice, which is not whether to use it but where, exactly, in the process. AI is not one decision. It is a tool that belongs at some stages and is poison at others.
Generate the ninety, judge the ten
The shape of every project is the same. The first ninety percent gets you a competent result. The last ten percent makes it good. AI has become extraordinary at the ninety and remains useless at the ten, because the ten is judgment applied at a scale too small to specify. Knowing where that line falls is the whole skill of using these tools well.
Spend generation freely. Spend judgment where it is actually scarce.
Our workflow, stage by stage
- Discovery and intent: human only. A model can summarize a brief; it cannot tell you what is worth making. This is where the work is won or lost, and it does not automate.
- Divergence: AI heavy. Generate many directions fast, far more than a person would have the patience to sketch. Breadth is exactly what the machine is for.
- Selection: human only. Choosing which direction is right sets the ceiling for everything after it. Hand this to the machine and the work goes generic on the spot.
- Production: AI assists. Drafts, variations, boilerplate, the hundredth state of a component. Let it carry the volume while you watch the quality.
- The finish: human only. The spacing, the timing, the one word that was almost right. The last ten percent is taste, and taste is the part that does not generate.
The trap: letting AI make the selection
There is one mistake that turns AI from leverage into rot, and it is subtle. It is letting the machine make the selection decisions, not just the generation ones. Generation is choosing from a blank page, and automating it is pure upside. Selection is choosing between options, and it is where your taste actually shows up. Outsource that and you have not sped up the work, you have removed yourself from it, and everyone can feel the absence.
What changed, precisely
AI did not replace the designer. It moved the bottleneck. For my whole career the constraint was output: making the thing took time, so that is where attention went. Now output is cheap and the constraint is judgment, so that is where attention has to move. We did not adopt AI to make less. We adopted it to spend our scarce attention on intent and finish, and let the machine have the middle.
Asked & answered
Use it for breadth and speed: divergent exploration, variations, first drafts, and commodity production. Reserve human judgment for intent, for choosing between options, and for the final ten percent that decides whether the result feels authored.
Only if you let it make the selection decisions. Generation can be automated; choosing which option is right, and finishing it, is where taste lives. Keep those stages human and the work stays distinctive.