Here is how I actually use the thing I built, and it is less impressive than you’d expect: I tell it what’s on my mind, and that’s it.
Not a structured request. Not a role and a task and a format. Just the stuff that’s rattling around — the thing I’m worried about, the thing I keep forgetting, the decision I’ve been circling for a week. Then I go do something else.
That’s the whole interaction model, and it’s the one the product is built for. Everything below is why.
Prompting is a symptom, not a skill
There is an entire economy teaching people to prompt — courses, templates, a hundred ways to phrase a request so a model behaves. I think that’s a symptom of unfinished products rather than a gap in users, and I’ve written the argument out properly.
The short version: if the product needs you to phrase things a particular way, the product is asking you to do its job. My grandmother should not have to learn a syntax to get help with her week, and a thirteen-year-old should not need a course to be understood.
So the design target isn’t a better prompt box. It’s a system where saying it badly still works.
Say it messy
The thing people get wrong at the start is trying to write a good request. They over-specify, they clean it up, they think about wording. All of that is wasted effort, and worse, it strips out the context that would have been useful.
Compare:
“Draft a professional email to a client about a project delay.”
with:
“the deck slipped again, third time, and I still haven’t told them — I don’t want to sound like I’m making excuses”
The second one is worse writing and a far better input. It carries the history, the stakes, the tone you’re worried about, and the actual problem — which is not the email, it’s how to say something true without it landing badly.
Give it the second kind. The mess is the context.
Say it in pieces, over time
You don’t have to fit a thought into one message. Several things over a morning — the appointment you need to move, the person you keep meaning to call, the thing you decided at lunch — all land in one place and stay there.
This is the difference memory that composes makes. Nothing important dissolves into chat prose: it becomes a memory, a plan, a reminder, a report, or an explicit decision that it isn’t worth acting on. So a fragment you mention on Tuesday can still be doing something on Friday, without you re-explaining it.
Chat assistants make you carry the thread. This one carries it.
Then leave
The part that takes the longest to get used to: you don’t have to stay.
The baseline state here is working, not waiting. You appear again at the edge that matters — hit send, submit, buy — and what arrives is finished work with one thing to approve, rather than a to-do list handed back. If Aura says it’ll do something, there’s scheduled work behind the sentence, and it’s done when it’s verified done, not when it was delivered.
So the correct move after dumping your thoughts is to close the app. That still feels wrong to me some days, and it’s the right instinct to fight.
What you never have to do
Pick a model. You’ll never see a picker — that’s deliberate, and the case for it is that choosing is our hardest job, not yours.
Learn a syntax. No commands, no magic words, no format.
Repeat yourself. If you told it once, it’s held. If it got something wrong, correct it in ordinary words — memory is yours to see and to fix.
Perform for it. You can be blunt, half-asleep, or in a bad mood. It isn’t scoring you.
The one thing worth doing deliberately
Be honest with it about what actually matters.
The quality of what comes back tracks the truth of what went in — not the polish. “I should probably exercise more” produces a shrug; “I’ve stopped sleeping properly since the job thing and I know they’re connected” produces something useful. Same effort, completely different input.
That’s the only skill I’d claim exists here, and it isn’t a prompting skill. It’s just being straight about your own week.