Lesson 1 of 6
Designing for probabilistic output
6 min read
Type the same request into an AI twice and you get two different answers — both fine. The software you grew up on never did that. So what does 'correct' even mean now?
No single right answer
Classic software is deterministic: one input, one correct output, every time. Generative AI is probabilistic — ask for a tagline, an email, an image, and there are dozens of good answers and no single right one. Designing as if there's one 'the answer' fights the material.
When output is probabilistic, the interface's job isn't to deliver the answer — it's to offer good options and let the person choose.
Design the choice, not the verdict
So show the work as a set of candidates, not a single verdict. Offer variants side by side, make Regenerate a first-class button, and let people edit what comes back. The best AI products treat every output as a draft the user steers — not a ruling they must accept or reject.
Variants, regenerate, and edit turn 'take it or leave it' into 'pick and refine' — the interaction probabilistic output actually wants.
This reframes failure, too. If one output is weak but the next is great, a variant picker quietly absorbs the miss — where a single-answer UI would just look broken.
The shape of it
- —AI output is probabilistic — many good answers, no single right one.
- —Offer variants and a Regenerate button instead of one fixed verdict.
- —Treat every output as a draft the user picks from and edits.
You're designing a feature that writes product descriptions. What's the most AI-native interaction?
Continue in the app
Take the whole Designing AI UX (for uncertainty) course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.