The Phone Can Stay
AI can take over the correspondence that keeps life running. The conversations we want to have will still be ours.
Imagine you’re at dinner with a friend who’s about to tell you why she left her job. Your phone is on the table. By midnight, a warranty claim on your broken dishwasher has to be filed. You know there’s a receipt somewhere. You pick up the phone, and your friend waits while you find out whether the serial number is on the receipt or on the machine.
The same rectangle holds two very different demands on your attention. You could use it tomorrow to call your friend and ask how she’s doing. Tonight you’re using it to satisfy a deadline set by an appliance manufacturer. A report might count both as communication. Your friend needs you to listen. The manufacturer needs the right information by midnight.
My forecast is that the smartphone, or whatever we eventually call the pocket-sized mobile device, will remain the dominant personal communications device for the foreseeable future. AI will change its job. We’ll spend less of our time exchanging messages to get routine work done, and we’ll talk with other people more often than we do today. To understand why, start with an intelligence you already rely on all day without ever having to open a conversation with it.
Your first subconscious
Stand up from a chair. Briefly, gravity pulls blood toward your legs. Pressure-sensing receptors and autonomic reflexes help adjust circulation so that blood keeps reaching your brain. You can cross the room in the middle of a sentence without giving any of that work a thought. If you feel light-headed, you might sit back down and give it your full attention.
There’s plenty of signaling inside your body. You experience the effects of that signaling as a person who can stand up and carry on talking. That’s what I mean here by the subconscious. I’m using an everyday word for the enormous amount of work that happens outside deliberate awareness. It includes ordinary regulation and learned routines that would make life unmanageable if each step needed a conscious instruction.
Some models in neuroscience take the analogy further. Anil Seth and Karl Friston describe how forecasts about the body’s internal state could guide regulation. Think of how a body can get ready for a demand before it reaches a crisis. A useful prediction lets a system act in advance. Your attention is available for whatever you’re doing while the adjustment takes place.
A personal AI can become a second subconscious for the part of your life that takes place in institutions. Its domain is higher-order in a practical sense. It works with promises, dates, permissions, and the plans you form with other people. It knows a receipt belongs to a dishwasher, the dishwasher belongs to your household, and the household has a claim that will expire on Tuesday. A prediction gives it somewhere to start. Authority and access let it do something about what it anticipates.
We can already give an AI instructions in conversation. For recurring work, the point of that conversation is to tell it how to carry the work the next time. You can set a preference once and update it when your life changes. Imagine how quickly you’d give up on a second subconscious that wanted to talk you through every adjustment.
A receipt for the work
At installation, you authorize your AI to keep the dishwasher receipt and serial number and file a standard claim if the machine fails. Later, you mention the fault or the machine reports one. The agent matches the fault to the right purchase, checks the warranty terms against the actual document, gathers the required evidence, and submits through an authorized channel. It saves the manufacturer’s acknowledgment, tracks the case, and follows up until the claim reaches a real endpoint.
A reminder would still leave you to file the claim. A drafted letter would leave you to send it and find out what happened. A submission number proves the claim reached the intake system. A confirmed repair appointment proves someone agreed to come. The assistant has to know what each receipt establishes because it determines what still needs doing.
The better this works, the less often you have to think about the AI. You can still inspect what it did, correct its records or revoke its authority. When it genuinely needs you, the interruption should tell you what only you can decide.
- DraftThe request is ready to send.
- IntakeThe manufacturer received the claim.
- AppointmentA technician is scheduled to come.
- RepairThe dishwasher works again.
What makes the work disappear from your evening is reliable access to the parties that can do something, a limit on what the AI can agree to, and a way to see that the job really happened. NIST’s work on agent identity and authorization addresses part of that problem. A manufacturer needs to know who is acting, on whose behalf, and with what permission. A capable model behind an inaccessible portal will still leave the work on your table.
What the timeline actually shows
The consultant’s slide usually goes like this. Telegraph. Telephone. Fax. Email. SMS. Smartphone. Put an arrow after the smartphone and ask what AI will put there. It’s an attractive exercise, especially for a company looking for a new piece of hardware to sell. Look closely at what each entry gave people and the arrow loses its force.
- Telegraph 1844
- Morse’s Washington–Baltimore demonstration showed that words could arrive over a wire far ahead of the person carrying them. The recipient got a message.
- Telephone 1876
- Bell’s early demonstration carried speech. People at a distance could hear each other and reply in the same exchange.
- Fax 1861 artifact
- Caselli’s pantelegraph could reproduce handwriting and drawings across a wire. That early machine predates Bell’s demonstration, even though the office fax boom came much later.
- Network email 1971
- Ray Tomlinson made it possible to send messages to users on other ARPANET computers. The addressee could read and reply later.
- SMS 1992
- The first SMS on Vodafone’s network was sent from a computer to a mobile phone. Short, asynchronous messages could reach a device you carried.
- Smartphone 2007 milestone
- Apple introduced the iPhone as a phone, media player, and internet device in one. Today’s smartphone holds phone calls, texts, email, and video.
Some entries change what a message can carry. Others change when or where it can reach someone. The smartphone bundles channels that already existed. A telephone call lives inside that bundle, so one of the oldest entries on the slide is also a feature of its newest entry. The history gives us a set of solved problems, and each had its own reason to catch on.
Now ask what problem the AI is solving. If you already have a place to read and answer messages, more convenient access can help at the margin. A warranty claim imposes a different cost. It needs an owner who will remember the deadline, find the records, act, and see the case through. The second subconscious can take responsibility for that sequence. Whether you heard about the claim through an earbud or saw it on a display mattered far less than who had to finish it.
There will be new ways for software to exchange information. Let the household agent and the manufacturer’s system agree on a repair slot through a shared protocol. Your friend can still reach you on the same number. AI can even translate her words as you speak, extending what a call can do with both of you still in it. A great deal can change behind an ordinary conversation.
The strongest case for glasses
Mark Zuckerberg gave the wearable argument a clear form in a 2025 interview with Ben Thompson. He sees glasses becoming the phones of the future. Worn on the face, they can give an assistant access to what you see and hear and let it respond in the moment. Try asking for the name of a bird while holding a child’s hand. Try getting live spoken translation in a market, or getting a description of a street sign you can’t read. Those are good reasons to put some computing in a pair of glasses.
Google’s November 2026 eyewear announcement is a concrete example. Its new glasses offer hands-free assistance and translation, and Google lists compatible smartphones for them. Meta’s Ray-Ban Display brings existing messaging services into the lens. An earbud helps with private audio. A headset can give a distant class a shared sense of space. A pin offers another place to put a microphone and camera. Some people will reach for these first for particular tasks, especially where accessibility or hands-free use matters.
Replacement would take a broader change. The pocket device lets you type a difficult sentence silently, read a long reply, show somebody a photograph, make a video call, authenticate a purchase, and put the whole thing away. It already fits the contacts, services, and accessibility tools that people depend on. In Pew Research Center’s 2026 survey of 5,511 U.S. adults, 91% reported owning a smartphone. That tells you how many adults in one country already have the device a successor would have to displace. It’s also where a personal AI can offer to do its work today, with no new hardware purchase.
The industry itself has doubts about a quick takeover. In a 2025 WPP Media survey of industry experts, half of the 62 respondents rated it highly unlikely that sales of AR glasses, VR headsets, and smart earbuds together would exceed sales of smartphones and smartwatches by 2030. That survey captures expert opinion about unit sales. For this forecast, I’d watch whether people stop carrying a phone on ordinary days and reliably do their essential communication on a different device. An eyewear sale tells me a lot less than a week of leaving the phone at home.
What the grocery store changed
For someone who must grow or hunt for dinner, obtaining food can organize the whole day. For someone with a stocked grocery store nearby and the means to shop there, bringing home food can take one stop on the way back from work. Someone still plants, harvests, packs, or drives. People also still hunt and grow food for livelihood, subsistence, tradition, and pleasure. The change is that many households can eat without carrying the whole process themselves.
USDA researchers trace a version of that shift in the United States. Between 1948 and 2017, agricultural output nearly tripled while agricultural employment fell from 13% to 2% of U.S. employment in their measure, which also includes forestry, fishing, and related industries. Machinery, specialized labor, and institutions made the store possible. I expect dependable AI to shift a similar burden of coordination. The claim still has to go through. It becomes conceivable that no one in the household spends dinner figuring out how.
I think of human beings as divine. By that I mean our capacity to love, create, grieve, teach, and give one another our presence is worth protecting. Ordinary administrative work can take skill and care. We do it because a life has to be maintained. We can value that work while making it possible for people to spend less of their one evening doing it.
More time for each other
Fewer administrative messages and more human conversation can happen on the same day. An assistant can quietly find a time when two friends are free, check the travel time, and handle the reservation. A friend now has an easier decision to make: do I want to see this person Thursday? Someone still has to make the invitation theirs. Once they’re together, the jokes, awkward silences, and changes of heart take exactly as much human participation as they ever did.
Early studies let us see the first part of the mechanism. In a staggered rollout studied by Erik Brynjolfsson, Danielle Li, and Lindsey Raymond, customer-support workers with AI assistance resolved 15% more issues per hour on average. A randomized field experiment by Eleanor Dillon and colleagues found that an AI tool changed email time more readily than meeting time. Those studies measure work and time. What people do with time they recover is another question. An employer can fill it with more tickets, and a platform can fill it with more prompts.
I expect more human contact because some of the calls we already want to make keep getting crowded out. And the medium can matter. In experiments by Amit Kumar and Nicholas Epley, exchanges that included voice created stronger feelings of connection than text in the situations they studied, without making the exchanges feel more awkward. The humble phone call has room to grow even in an age when companies can put a display in your glasses.
For a way to test the forecast, count the calls to friends you keep putting off, the visits you actually make, or the evenings when everyone can finish a story. Count the claims that were resolved and the amount of attention their owners had to give them. The volume of machine messages could grow considerably while the human burden of corresponding about routine matters shrinks.
Back at dinner
The repair is booked for Tuesday. If you turn the phone over, you can find the confirmation. Your friend is still telling you about the day she decided to leave. There’s a moment when she stops and looks down at her glass. You wait a beat because you know her. Then you ask how it felt to walk out.
Try it in class
Take a recent thread about rescheduling a dinner, getting a refund, or arranging a repair. Identify the moment that required someone’s genuine judgment. Give an assistant the remaining work, specify what authority it would need, and name the evidence that would let it close the task.
Now propose a new wearable. Describe the one thing it makes easier, then list what people would have to do with it before you’d count it as their primary communications device. What evidence would change your mind?
1 thought on “The Phone Can Stay”