The Feed Turns Around

Those two screens are not two styles of design. They are two answers to one question: who moves first?

A box is the interface of a tool that waits. A hammer waits. A search engine waits. The chat apps we all learned on over the last few years wait. Nothing happens until you type, and everything that happens is a reply to you. The box is honest about it. It is empty because nothing has happened yet.

A feed is the interface of a world that moves without you. The first one was a machine. In 1867 Edward Calahan unveiled the stock ticker in New York, a telegraph that printed trades onto a ribbon of paper as they happened. The market did not wait for a broker to ask. It moved, and the tape was how a person kept up with something that would not stop moving. Every feed since, the wire service, the inbox, the timeline, is the same tape: events you did not request, in the order they happened, newest on top.

That is the principle. Interface follows initiative. If you start, you get a box. If it starts, you get a feed. No amount of design talent changes this, because a box cannot hold events nobody asked for, and a feed cannot hold a question nobody has asked yet.

Why the box is about to break

Look at what shipped this fall. In the space of a few months, nearly every major AI company released a personal agent that is always on: OpenAI’s Dots, Meta’s Muse, Anthropic’s Claude Duet, SpaceXAI’s Grok Bot, Google’s Gemini agent, with Amazon’s Alexa already sitting on the kitchen counter. The whole point of these products is that they move first. They watch your inbox, your calendar and your accounts, and they act while you sleep.

An agent that moves first produces events, and events do not fit in a box. You can already see the strain. The early versions keep a running list of what they are working on and what they need from you, tucked in beside the chat. That list is a feed waiting to be promoted to the home screen. Google has already started down the road. It is quietly retiring its Spark brand and folding the agent into a simple switch under the prompt: Chat or Task. That switch is the box admitting it has two jobs.

Here is the part most people will miss. The feed comes back with the same scroll, the same thumb and the same tiny taps we spent fifteen years learning on social media. But it turns around.

The social feed points outward, and your attention is the crop. The activity feed points back at you, and your attention is the steering wheel.

On a social feed, you scroll through other people’s lives so that someone can sell your looking. On an activity feed, you scroll through your own errands, bills, appointments and replies, and every glance is a small act of government over a machine that is living part of your day for you.

IRL and SIM

Every item in that feed will carry one of two tags.

IRL means it happened. The bill is paid. The dentist moved to Thursday. The reply to the landlord went out.

SIM means it has not happened, but it could. It is your AI showing you work it would take off you if you let it. Sometimes that is a small decision: I would have declined this vendor demo. Once it knows you, it is a whole job: your traffic hearing is on the 10th, and I would prepare all of it. Under every SIM card sits one button. Try it.

Nobody writes those cards by hand. Your AI builds them by noticing the patterns of your life: your email, your texts, your calls, the places you go, the things you keep asking it for. Anything you do the same way often enough becomes a candidate.

SIM is the most important idea in the whole design, because it breaks the loop that has stalled delegation forever. You cannot trust someone with work until you have seen them do it, and you cannot see them do it until you trust them. Software engineers broke that loop long ago with what they call shadow mode or a dark launch: the new system runs alongside the old one, makes its decisions on real inputs, and nothing it decides is allowed to touch the world. You compare. When the shadow has been right long enough, you switch it on. Pilots do the same in simulators. Traders do it with paper trades.

SIM is shadow mode for your life. The feed is really two feeds woven together: a record of what your AI did, and a running audition for what it wants to do next. Tap Try it, and a little more of your life moves from the SIM column to the IRL column, for good.

Notice what you never had to do. You did not imagine the job, describe it, or build anything. You did not stare at a box and think of a prompt. The AI noticed the load, sized it, and offered to carry it. Your whole part is agreeing. That is what turns a personal AI from a tool you use into an ambient partner you barely see: less to remember, fewer decisions, less weight on the day.

What I expect

1. The empty box stops being the home screen.

   Within two years, the major personal AI apps will open on a stream of activity, not a blank prompt. Chat survives as a button in the corner, the way search survives inside apps that are mostly feeds.

2. Every action carries a reality tag.

   The words may not be IRL and SIM, but the split will be universal: a visible line between what your AI did and what it would have done. The shadow card becomes the standard way an agent asks for more authority, because asking in the abstract never works and showing does.

3. The tap becomes a signature.

   The gestures we learned on social media get rewired from reaction to permission. A like meant “I saw this.” Try it means “do this from now on.” Swiping away a SIM card is a standing no. People will discover, a little uneasily, that their thumb has been signing things.

4. The feed is built to shrink.

   Social feeds are paid to grow. The activity feed is paid to disappear. Routine actions will collapse into a single line, fourteen things handled, and the best personal AI will be the one whose feed you barely need to open. Companies will compete on how little of your attention they take.

5. The feed becomes the record.

   When something goes wrong, a payment, a promise, a message you never would have sent, the feed is the evidence: what was done, when, and on whose authority. People will screenshot their AI’s feed the way they screenshot texts today. And the SIM column will become an oddly honest portrait of who you are, drawn from what you consistently say yes to.

I am living a crude version of this already. In September I told my own AI, in plain words: you decide. Since then, much of what reaches me is not a question. It is a list of things already done, and my job is to scroll it and redirect when something is off. I wrote about the practical side in You Do It. You Decide., about what finished work feels like when it arrives in The Notification Everyone Wants Is Done, and about why control never required matching the machine move for move in Two Buttons Are Enough. Try it and swipe away are those two buttons. What I did not see until now is the shape of the whole thing. A list of things done for you, newest first, that you govern with your thumb, is a feed. I just have not had a SIM column. I have been running the shadow by hand, one redirect at a time.

It is also the argument of The Zero-Attention Economy arriving at the one screen people look at most. The feed was the great machine for capturing attention. Turned around, it becomes the place where attention is spent only on the exceptions.

How this could be wrong

Already true

Every major AI company now ships a personal agent that acts without being asked, and at least one already keeps a running list of its own work beside the chat. Shadow mode is ordinary practice for deploying software that makes decisions. And billions of people already know how to govern a feed with one thumb.

What has to happen

Agents have to do enough on their own that the volume outgrows a chat thread: dozens of actions a day, not a few. And they have to be allowed to earn authority in small, specific grants rather than through all-or-nothing permission screens. That is a design and trust problem, not a model problem.

Where I am probably wrong

The feed might be skipped entirely. If voice wins, you will not scroll anything; your AI will tell you three things over coffee and that will be the whole interface. Or the companies will notice that a feed is a familiar place to sell attention and refill it with suggestions, offers and bait, turning the activity feed back into a social feed with my name on it. If that happens, I will have called the right shape and the wrong economics. And SIM needs restraint. An AI that shows you forty auditions every morning is not delegation. It is a new kind of spam.

Go back to Sunday morning and the empty box asking how it can help. It was never going to last. A box is a promise that nothing happens until you speak, and the whole point of a personal AI is that things happen anyway. The next time you open it, it will already be full. Not with other people’s lives. With yours.

It started without you.

Part two follows the money: Try It Is the Business Model. Part three shows you how to build your own: Build Your Own SIM Feed.