Ask Them. Ask Them Every Time.

In 2010, on stage at D8, Steve Jobs was asked about privacy. He didn't reach for a policy or a framework. He said this:

"Privacy means people know what they're signing up for, in plain English, and repeatedly. Ask them. Ask them every time. Make them tell you to stop asking them if they get tired of your asking them."

Sixteen years later I still think that is the most complete specification of a consent system anyone has written. Four sentences. No committee. And almost nobody built it, including — for a long stretch — us.

I want to take it literally, clause by clause, because each one turns out to be a hard engineering problem wearing a very simple sentence.

"In plain English"

A person should never see prefs.food.allergies.

That sounds obvious and it is not what the industry does. Every consent system I have used asks in the vocabulary of the system rather than the vocabulary of the human. A cookie banner with 43 vendor toggles is technically informing you. It is not asking you.

So we built purposes. A hotel concierge does not request eleven dot-paths; they ask *"may we look after you during your stay?"* — one question, in the language a person actually thinks in, bound underneath to exactly the fields that question needs. A field-service technician asks *"may I come to your home and get it right?"* A billing team asks *"can we sort out your bill?"*

Fifteen of these are published now. The person reads a sentence. The machine reads the scopes. Neither has to learn the other's language.

"And repeatedly"

This is the clause everyone skips, and it's the expensive one.

Repeated asking is only tolerable if each ask is cheap. Ask badly and often, and you have built the cookie banner — which is *technically* Jobs's design and is universally hated, because the asking costs the person more than the answer is worth to them. Consent fatigue is not a failure of will. It's a failure of interface economics.

The cost of consent is paid in interactions. So the engineering job is not "ask less". It's make the ask so cheap that asking often is a kindness rather than an assault. One sentence, two taps, expiring by default, and never the same question twice in a week.

Every purpose we publish carries a TTL. A dinner reservation's grant lasts seven days. A hotel stay, fourteen. Your privacy stance, a year. Nothing is forever, because forever is how "ask them repeatedly" quietly becomes "ask them once and never again."

"Make them tell you to stop asking"

The part that requires actual humility: the system has to accept *no* as a durable answer and stop.

Not "no for this session." Not "no until we redesign the modal." Revocation in our system is immediate and fabric-wide — no notice period, no grace window, no thirty days. And a person who has told us to stop should not be asked again by the same party for the same thing.

That's the clause that costs revenue, which is presumably why it's the clause nobody implements.

What would Apple do?

I get asked this a lot, and I think the honest answer is: Apple would say no to more things than we have.

The company's actual method — the one visible in the products rather than the keynotes — is subtraction under constraint. Not "what else could this do" but "what must be true, and what can we therefore refuse?"

Applied here, I think it produces three refusals.

It would refuse to make everything a market. We publish 263 things a person can share. 153 of them can be offered at a price the person sets, which I believe is the correct inversion of an industry where you are the inventory. But eighteen of them — a journal, therapy notes, what you have come through, how you have been treated — cannot be auctioned at any price, by anyone, ever. Not as a policy. As a property of the code, with a test that fails the build if it changes.

We didn't get that right first. We published those scopes routed to a plaintext store while our own architecture is encrypted end to end. Nothing leaked — there is no way to write them yet — but the design was wrong, and we caught it by going looking rather than by being told.

It would refuse to be clever about identity. A brand that reads your preferences never learns who you are. Each one receives a different pseudonym for the same person, so two of them comparing notes learn nothing. We got *that* wrong too, and shipped the raw identifier for a while. It's fixed, and there are six tests pinning the properties, including one for a delimiter collision that would have made two different people look like the same person.

It would refuse to ship the demo. The thing I keep coming back to about Apple is that they don't announce the version that works on stage. We publish 263 scopes and our server currently binds a fraction of them. Anyone can check — the registry is a public URL — and until that's closed I'd rather say it plainly than let someone discover it by building against us.

The part Jobs didn't have to solve

There's one thing that specification doesn't cover, because in 2010 it didn't exist: the asking is no longer human-to-human.

Increasingly the entity that wants to know something about you is an agent. It works for a shop, a clinic, a bank, an insurer, a field-service dispatcher. It is fast, it is tireless, and it will ask a thousand people the same question before lunch.

Jobs's rule still holds, but it needs an addition: if agents are going to do the asking, then a person needs an agent of their own to do the answering. Otherwise "ask them every time" becomes a denial-of-service on human attention, which is exactly what happened to the cookie banner.

That's what 🤫 Agent One is for. It is your side of the conversation. It knows the answers you've already given, it can answer routine questions without waking you, and it brings you the ones that genuinely need a person.

And here's the part I find genuinely delightful: an agent that holds your preferences and your consent is not antisocial. It's the opposite. It can *introduce* you.

Two agents can establish, without either human being disturbed and without either side learning anything they weren't granted, that there's a reason to talk. A local specialist who does the exact thing you need. A shop that actually has your size. A person in your community working on the thing you care about. The agent's job is to find the meeting worth having and then get out of the way — because the point was never the agent. The point is two people who should have met, meeting.

Private and social are not opposites. A person who is confident about what they're revealing is far more willing to be known. The reason people are guarded online is not that they hate connection; it's that they have no control over what a connection costs them. Give them the dial and they turn it up, not down.

Where we actually are

Published and live: the registry, the purposes, the tiers, the receipt chain, the deploy pipeline that verifies every release against the running site before promoting it.

Not done: the server binds a fraction of what we publish. There's no interface for a person to fill in the deeper parts of their profile. The encrypted path for the sensitive tier isn't built. Money moves on exactly one rail.

I'd rather write that list down than let it be discovered. It's the same instinct as the receipt: show the proof, or don't make the claim.

Jobs was right in 2010 and the industry spent sixteen years proving it by counterexample. Ask them. Ask them every time. In plain English. And when they say stop — stop.

The rest is engineering.

Sources and further reading

• Steve Jobs on privacy, D8 Conference, June 2010 — the quotation above is from his onstage remarks, widely reported and available in the conference video.
• The live scope registry — 263 scopes, 15 purposes, versioned and public
• The Supercomputer in Your Pocket Wants a Job — the technical companion to this piece
• We Published Forty-Seven Scopes and Built Eight — the drift, and the fix
• What we have not done — the standing honest-gaps page

*Apple is named here because its public design philosophy and Steve Jobs's public remarks are the subject of this piece. Apple is not a partner, customer, shareholder, or investor of HushOne, Inc. or Hushh Technologies Corporation, and has not reviewed or endorsed any of it. Our own compliance certifications are in pursuit and not held.*