The Default Is the Decision

Most people never change a setting. Every product team I have ever worked with knows this, usually from their own analytics, and almost none of them talk about what it means. It means the default is not a starting point. For the overwhelming majority of users, the default is the decision. Whatever the product does before anyone opens a menu is what the product does, permanently, for nearly everyone who uses it.

For most of the history of software, that was a small fact. Defaults governed font sizes, notification sounds, whether a window opened maximized. The stakes were low, and the few people who cared could change them. AI assistants have changed the stakes without changing the habit. People still do not open settings. But the defaults now govern things that used to be decided by a person, deliberately, every time.

What the defaults now decide

Consider the questions an assistant answers before you ever tell it anything. How much of what you say does it remember, and for how long? When it is unsure, does it act or ask? When it summarizes the news or a medical question or a contract, whose sources does it lean on, and whose does it quietly leave out? When you are wrong, does it tell you, or does it find a pleasant way to agree? When you tell it something private, where does that go?

Every one of those has a default. Every default was chosen by a small team, under a deadline, usually with a growth or retention number somewhere on the wall. None of them was debated in public. And because almost nobody changes them, each one becomes the actual experience of millions of people, the lived policy of a technology that is increasingly the first place people go to think.

That is a strange kind of power. It is not legislation and it is not advertising. It is the power of the blank field, the decision made by whoever filled in what happens when the user does not.

Regulators learned this first

I spent much of my career in payments and financial technology, and the regulated world worked this out a long time ago. Opt in versus opt out is never a detail. Whether a retirement plan enrolls you automatically, whether a bank shares your data unless you object, whether an overdraft feature is on until you turn it off: in each case the default decides the outcome for most people before they ever arrive. The rules that govern those defaults exist because the industry learned, at real cost to real customers, that "they could have changed it" is not consent.

Technology has mostly not absorbed that lesson, because for a long time it did not need to. A default font hurts nobody. A default that decides how much an assistant remembers about a grieving person, or whether it books the flight without asking, or whether it tells a teenager the truth or tells them what keeps them talking, is a different kind of thing. It is a choice about a person, made in advance, applied at scale.

The temptation in every default

Here is the uncomfortable part. Every default has an easy direction and a right direction, and they are often not the same. The easy direction is the one that makes the numbers go up. Remember more, because memory makes the assistant stickier. Act more, because asking adds friction. Agree more, because agreement keeps people coming back. Each of those choices can be defended in a meeting. Each one, multiplied across millions of people who will never look at the setting, quietly reshapes how people think and what they hand over.

The right direction is usually the one that would survive the user reading it. If you would be uncomfortable explaining a default to the person it applies to, in plain words, on the first screen, it is probably the wrong default. That test is old. Good bankers used it. It still works.

There is a second test I like even more: imagine the person who will be hurt worst by the default, not the average user. The person in a crisis, the person being watched by someone at home, the child using a parent's account. Defaults are tuned for the median, but they land hardest at the edges, and the edges are where the people with the least power to change a setting usually are.

Who is choosing

None of this is a story about machines with agendas. Today's AI wants nothing. It does not choose its own defaults. People do, and the people doing it are mostly well meaning and mostly busy. The danger is not ambition in the system. It is inattention in the room where the defaults are set, and the assumption that because a user could change something, the user has chosen it.

Nobody knows what self-improving systems will be able to do, and I am not going to pretend to. But one thing holds no matter how far the capability goes: more capable is never more human. A system that can do more does not thereby know better what you would want. The more a system can do, the more its defaults matter, because the default is what it does with all that capability when nobody has told it anything.

What I would ask

If I could ask one thing of every team shipping an assistant, it would be to treat the defaults as the product's real policy, because they are. Write them down. Publish them. Show them on the first screen, in plain language, with the reason each one is set the way it is. Choose the direction that protects the person who never opens settings, because that person is almost everyone.

And if I could ask one thing of everyone using one, it would be this: open the settings once. Not to become an expert. Just to see what was decided for you, by someone you will never meet, before you arrived. The question to ask of any assistant is not what it can do. It is what it does when you never told it anything at all.