AI prompting patterns
Notes on prompts as design objects: structure, tone, voice, and the difference between an answer and a useful answer.
- Stare
- Studiu
- Mediu
- Text, LLMs
- Actualizat
- aprilie 2026
Prompts are interfaces. They have hierarchy, voice, defaults, and edge cases. I have been treating them with the same care I would treat a form, a settings page, or a piece of UX copy.
This track is a collection of patterns that keep showing up: how to scope a task, how to set a tone, how to give the model just enough context without drowning it.
01 · Note
Prompts as products
A prompt is a product surface. Someone is going to copy it, edit it, paste it into a context window, and live with the answer for a while.
That means the same questions apply: who is this for, what does success look like, what is the failure mode, what gets cut.
02 · Note
Voice and defaults
I am especially interested in how the framing of a prompt sets tone. A small change in voice up front tends to do more for output quality than long lists of rules at the bottom.
Defaults matter too. If I want short answers, I have to say so early, and ideally show what short means.
03 · Note
Plates
Întrebări deschise
Open questions
- What is a prompt's equivalent of an empty state?
- How do you version prompts the way you version components?
- When should a prompt say less, not more?
Dacă simți că ceva nu e în regulă, începe de acolo.
Spune-mi ce se întâmplă și cum ar arăta un rezultat bun. Un email scurt e de ajuns — le citesc pe toate și răspund în două zile lucrătoare.