Look up next time you’re on a busy street. Heads down, thumbs moving, people steering around each other by peripheral vision and luck. We’ve spent almost twenty years since the iPhone learning to walk, eat and hold conversations while staring at a rectangle. My prediction: from 2027, that starts to reverse. The future is screen-less, not because screens disappear, but because AI agents will do most of the tapping for us.
The short answer
- A screenless future means you ask for an outcome and an AI agent does the clicking, through apps, APIs and connectors, with nothing for you to look at.
- Our time online is already substantial. DataReportal’s 2025 report gives 6 hours 38 minutes a day for adult internet users. That measures online time, not every screen. Survey age coverage changed in 2024, so it is not a clean year-by-year measure of whether screen use has peaked.
- The pieces are arriving. Amazon’s Alexa can already buy for you in the US, Alexa+ reached Australia in early access in August 2026, and you can text an agent like Instinct to book and buy.
- Trust and reliability both matter. Only 5% of Australian respondents were fully comfortable with AI completing a purchase independently, Adyen found in 2026. Permission to buy and the ability to finish correctly are separate problems.
- UI matters less, UX matters more. The experience moves from screens you navigate to tasks you hand over.
How much time do we spend on screens?
A lot, but the measures differ. DataReportal’s 2025 global report puts adult internet users at about 6 hours 38 minutes online each day. That is not total screen time, which can also include offline work, television and gaming. Its authors note that adding people aged 65 and over to the survey in 2024 affects comparisons with earlier years. The Australian weekly figure below comes from a separate report, so I would not treat the two as one trend line.
The iPhone was announced on 9 January 2007 and the iPad went on sale in April 2010. These devices changed how we use the web. The time-use reports below describe what people reported doing; they do not prove that AI has started reversing that behaviour.
- Globally, time online peaked at just under 7 hours a day in 2020 and 2021, then slipped. DataReportal’s Simon Kemp titled his 2023 analysis “The time we spend online is falling”. Mobile now takes about 3 hours 46 minutes of it.
- In Australia, We Are Social and Meltwater put us at 41 hours a week online, against a global average of 33.
- And yet we’re starting to pull back. Deloitte found Australians’ weekly social media time fell 16% in 2025. Its media lead said Australians are “learning to touch grass”.
The cost shows up in odd places. Ohio State researchers estimated that more than 1,500 US pedestrians were treated in emergency departments in 2010 for injuries related to phone use while walking, more than double the 2005 estimate. Humans weren’t designed to walk and read at the same time. We just got very good at it.
What is a screenless future?
A screenless future is one in which you ask for what you want by voice, text or glance, and an AI agent completes the task through apps, APIs, MCP servers and connectors without you navigating a screen. You still see a screen when it helps, to approve a payment or check a photo. You stop using one to do the work.
Here’s the difference in one sentence. Today you open an app, search for running shoes, compare, pick a size, enter an address and pay. In a screenless world you say “Alexa, buy me Nike runners in my size”, and the agent does the rest, then asks you to confirm.
That’s not science fiction. It’s a US product with an Australian version on the way:
- Amazon’s Buy for Me in Alexa for Shopping can complete eligible purchases from other retailers for US customers. In its second-quarter 2026 earnings discussion, Amazon said more than 350 million customers had used Alexa for Shopping in the preceding 12 months. Usage is not a count of autonomous purchases.
- Alexa+ in Australia opened in early access on 5 August 2026. It can check the price of a brand you prefer and will “confirm with you before ordering”.
- Instinct, the personal agent that raised US$1 billion at a US$10 billion valuation in September 2026, has no app to learn. Its own site says “You can text or call it.” I use it myself, over WhatsApp: no new screen to navigate, I just message it the way I’d message a person.
- Meta’s Muse works inside WhatsApp chats and can make purchases (US only for now).
- Australian banks have already run the payment side. In January 2026 Mastercard announced Australia’s first authenticated agent payments: an AI agent bought Event Cinemas tickets with a CommBank card and booked Thredbo accommodation with a Westpac card, with the cardholder’s consent at each step.
A caution on all of these: a launch describes an offered capability, not its reliability. The product announcements linked here do not provide a comparable, independently audited completion rate for real purchases. I would want to test the exact task, approval flow and failure handling before relying on one.
Underneath all of this is plumbing most people will never see. The Model Context Protocol, or MCP, is a standard way for an AI agent to connect to a business’s tools and data. In December 2025 it moved to the Linux Foundation’s new Agentic AI Foundation, with more than 10,000 published MCP servers and support in ChatGPT, Gemini, Microsoft Copilot and Claude. When the plumbing is shared, the screen becomes optional.
Which devices will replace the screen?
None will replace the phone outright. I expect speakers, glasses and small devices to take over more of the jobs we now do on a phone, because an agent does the work in the background. EssilorLuxottica reported more than seven million Ray-Ban Meta and Oakley Meta glasses sold in 2025, while the first standalone AI gadgets failed.
| Device | What it is | Where it stands, October 2026 |
|---|---|---|
| OpenAI and Jony Ive’s device programme | Hardware collaboration with Ive and LoveFrom; the io team merged with OpenAI in July 2025 | The primary announcement confirms the collaboration. Reported product forms and shipping windows should not be treated as confirmed retail launch dates |
| Amazon Alexa+ | Voice assistant that books, buys and manages lists | Live in the US; early access in Australia since August 2026 |
| Google Gemini for Home | Gemini on Google Home speakers and displays | Australian voice-assistant early access expansion announced in April 2026 |
| Meta AI glasses | Ray-Ban and Oakley glasses with cameras, speakers and Meta AI | More than seven million sold in 2025, according to EssilorLuxottica. Meta has announced plans to bring Muse to glasses |
| Meta Muse Charm | A pocket-sized voice device for the Muse agent | Announced in September 2026; no price or date yet |
| Humane AI Pin | A screenless wearable assistant | Switched off on 28 February 2025 after HP bought Humane’s assets for US$116 million |
The pattern I take from this: the first wave of AI gadgets tried to replace the phone before the agent behind them could do much. They lost. What’s working rides habits people already have: glasses they’d wear anyway, a speaker in the kitchen, a chat app they already have open all day. The device that wins will be the one you forget you’re using. A pocket device that knows your preferences and just gets things done is the obvious next step, and Meta has already announced one.
OpenAI hasn’t said much about its device, so I won’t speak for it. What Sam Altman did say in November 2025 is telling: using today’s devices feels “like I am walking through Times Square”, and he wants the new one to feel like “sitting in the most beautiful cabin by a lake”. That’s a company selling less screen, not more.
Does UI still matter if agents do the work?
Less than it did, and differently. When an agent completes the task, the buttons, menus and layouts people navigate matter less. UX gets redefined around handing over a task, trusting the result and approving the decisions that matter: a clear confirmation, a clear receipt and an easy way to undo.
I’ve spent more than 20 years building websites, so I don’t say this lightly. Good UI won’t vanish. It moves to the moments people still want to see: choosing a holiday, checking a design, approving a big payment. Everything routine moves behind the scenes. For a business, the interface that matters most becomes the one an agent uses: clear facts, a working enquiry or booking path, and tools an agent can call. That’s why I’m watching WebMCP, a proposed standard for websites to hand AI agents real tools, and why I think your website matters more, not less, in this world. I’ve made that case in full in why your website matters more than ever in the AI era.
Will people trust AI with their shopping?
Not yet. I think they will, the same way they came to trust online banking, but that is a view, not a certainty. Only 36% of Australians are willing to trust AI, and 64% are uncomfortable letting it complete a purchase. In 2000, 70% of US internet users feared hackers stealing their card number online. Today almost all Australian banking is digital.
When I was a teenager I bought the parts one by one and built my own computer. I remember the pieces spread across the floor, the screwdriver, and pressing the power button half expecting smoke. When it actually booted, my first thought wasn’t “this is amazing”. It was “how does this actually work, and can I trust it?” Could I put my files on it and expect them to be there tomorrow? Everyone was asking the same question back then. Shopkeepers wondered whether to move their ledgers off paper. Families wondered whether to put their accounts on a machine. Today nobody wonders whether their iPhone will wake them up. We set the alarm and go to sleep. The question was never really answered. It just disappeared. My bet is that AI crosses the same line. If it does, people will hand it their information and their shopping without a second thought. If agents keep making expensive mistakes, it won’t.
| Then and now | Figure | Source |
|---|---|---|
| US internet users worried about hackers stealing their card number, 2000 | 70% | Pew Research Center |
| Australian customer-bank interactions now on digital channels, 2025 | 99.3% | Australian Banking Association |
| Australians willing to trust AI, 2025 | 36% | KPMG and University of Melbourne |
| Australians uncomfortable letting AI complete a purchase, 2026 | 64% | Adyen Index 2026 |
| Australians fully comfortable letting AI buy on its own, 2026 | 5% | Adyen Index 2026 |
AI sits roughly where online card payments sat around 2000. It’s useful, slightly scary, and used mainly for low-stakes jobs. Trust gets built the same way it was then: small wins, clear receipts, the ability to reverse a mistake, and banks and card schemes standing behind the transaction. The analogy has limits: a card payment is one step you take yourself, while an agent makes several choices on your behalf, and a wrong one is harder to spot. But if people hand an agent their groceries and nothing goes wrong for a year, I expect they’ll hand it their holidays.
Where does the evidence push back?
The screenless future has failed before, and it’s still early. The Humane AI Pin was switched off within a year, OpenAI scaled back buying inside ChatGPT in March 2026, and the home robots in demo videos are often controlled remotely by people. The direction is right. The 2027 timing is a bet.
- Buying inside the chat stalled once. OpenAI said in March 2026 that the first version of its Instant Checkout “did not offer the level of flexibility” it wanted, and let merchants use their own checkouts instead.
- Agents raise real privacy questions. TechCrunch reported in August 2026 that Instinct’s terms gave it a “perpetual and irrevocable” licence over user material.
- Australia is behind the US. Most agent shopping features are US-only or in early access here.
When will robots do the chores?
I would not plan a household around a robot arriving in 2027. The evidence is a mix of company targets, manufacturing announcements and demonstrations. Those are different from a machine completing the chores in your own home. The useful checks are what it can do without help, when a remote operator steps in and what happens when a task fails.
- 1X NEO. In its April 2026 factory update, 1X said the first units would ship during 2026, with some customers receiving theirs later. It also described prioritising units for its internal team and home testing. A shipping target is not an independently verified delivery or task-completion result.
- Figure 03. Figure describes its robot as designed for homes and commercial applications. Its announcement explains changes to safety, hands, sensing and manufacturing. That design intent is not an independent test of everyday household autonomy.
- Tesla Optimus. Tesla’s 21 October 2026 shareholder update says production began in Fremont and labels the California programme an early ramp. Production status is not the same thing as a confirmed consumer delivery date or proof of reliable household work.
I’ll be honest about my interest here. I have two kids, and the laundry never ends. Most nights, once they’re asleep, there’s a mountain of clean clothes on the couch and I’m matching tiny socks that never seem to come in pairs. I spend hours a week folding. I run a company that builds with AI every day, and I still fold T-shirts by hand. I am desperately waiting for a robot that can do it. Folding is hard for robots because cloth never holds the same shape twice, which is exactly why it’s the job every robotics company keeps demonstrating. When one of them can fold my kids’ clothes unsupervised, I’ll be the first in the queue, and I’ll write about it.
What should businesses do now?
- Imagine your customer never sees your website. Could an agent still find your prices, hours, stock or availability, and book or buy? If not, that’s the gap to close.
- Put your facts in plain text and keep them the same on your site, your Google Business Profile and your listings.
- Make your main action work for software: a labelled form, a clear booking path, a confirmation an agent can read.
- Watch the connectors. If your booking, store or CRM platform offers an MCP server or agent connector, find out what it exposes and who can use it.
- Design for the moments people still want to see: the decision, the confirmation and the receipt.
I’m optimistic about all of this. Humans weren’t designed to stare at screens for hours every day. If agents take over the tapping and robots eventually take over the folding, we get some of that time back for the people in the room. For why the agents are arriving so quickly, read my piece on personal AI agents as the next big industry, and for the shift itself, what I mean by the agentic web.
Frequently asked questions
Will smartphones disappear?
Not soon. Phones will stay the place we approve payments, watch video and check photos. What changes is how much we do on them. As agents handle routine tasks by voice, text or glasses, the phone becomes the place we confirm and review, not the place we do every step ourselves.
What is the OpenAI and Jony Ive device?
It’s OpenAI’s first hardware, designed by Jony Ive’s team after OpenAI acquired his company io in May 2025. OpenAI hasn’t shown it or confirmed its form. Bloomberg reported a screen-free speaker with a camera, and a court filing said it won’t ship before the end of February 2027.
Can Alexa buy things for me in Australia?
Partly. Alexa+ has been in early access in Australia since 5 August 2026 and can check prices and order with your confirmation. The fuller “Buy for Me” feature, which purchases from other retailers too, launched in the US in May 2026 and hadn’t been announced for Australia as of October 2026.
What are AI wearables?
AI wearables are devices you wear, such as glasses, pins or pendants, that offer an AI interface. Meta’s glasses are a large commercial example, with more than seven million sold in 2025 according to EssilorLuxottica. Different form factors have different limits. A successful glasses product does not establish that every standalone AI gadget will succeed.
What does a screenless future mean for websites?
Websites become sources for agents as much as pages for people. An agent buying or booking for someone reads your site’s facts, prices and policies, and uses your forms or connectors. Sites that are clear, accurate and easy for software to use get chosen; sites built only for human eyes get skipped.
