Self-initiated concept
Atelier AI
A concept tool placing generative AI inside divergence, comparison, and convergence, focused on traceable design decisions rather than one-shot answers.
- Scope
- Product strategy, prompt system, experience and prototype design
- Duration
- Four-week concept sprint
- Tools / Methods
- Figma · Prompt systems · Generative prototyping
- Year
- 2026
A good generative tool does not choose the direction; it makes the reason for each choice clearer.
Product principle
The problem is not too little output, but lost judgment
Generative workflows quickly accumulate outputs but struggle to answer why one was kept or what changed next. Atelier AI breaks prompts into composable tokens and keeps selections, exclusions, and notes on one decision chain.
Diverge, compare, converge, trace
- 01
Diverge
Control variables
Each round locks stable tokens and changes only one or two directional variables.
- 02
Compare
View on equal terms
Consistent crop and scale reduce bias caused by presentation differences.
- 03
Converge
Mark reasons first
Keep and reject reasons are recorded before a choice enters the next round.
- 04
Trace
Preserve the chain
Any final output can be traced to its prompt, comparison set, and key judgments.
Keep the comparison surface neutral
The interface avoids misleading “best” scores and instead offers equal sizing, blind review, and custom criteria. AI expands the candidates; people still define what deserves to remain.
What the prototype clarified
The most important feature was not the generate button, but writing criteria before comparison. It turns instinctive reactions into discussable design decisions and keeps the next prompt round from becoming random tweaking.
Toggle tokens on the home page and watch curated outputs recombine deterministically.
Try the prompt-token machine