Connect with us

NEWS

Generative AI Is After the Hours You Still Live

Treat generative AI as a medium and the product map is clear: companions, walkable worlds, and the last private hours.

Published

on

Generative AI is already a medium, wrapping therapy, friendship and play in products. A 2026 study of 12,637 real uses put therapy and companionship first for the second year running.

Marshall McLuhan and Guy Debord, writing when television was the new box in the living room, left a test this moment still fits. Ask what still happens without an interface. Then watch which company builds one.

Debord’s Map of What Still Escapes the Market

An essay published Tuesday revived the two 1960s writers as guides to ChatGPT and Claude. The useful move is not the label. It is the inventory Debord made of life that had not yet been run through a medium, and therefore had not yet been sold.

In The Society of the Spectacle (1967), Debord wrote that modern life presents itself as “an immense accumulation of spectacles,” and that “everything that was directly lived has receded into a representation.” Thesis 4 is the line that still does work on a chatbot: the spectacle is a social relation mediated by images, not a pile of pictures.

When the real world is transformed into mere images, mere images become real beings, dynamic figments that provide the direct motivations for a hypnotic behavior.

Guy Debord, The Society of the Spectacle, thesis 18, 1967

He was watching television. The same test now reads as a product brief. Entertainment, exercise, sleep, friendship, romance and food were the leftover rooms. Mobile social software took several of them. Generative models are being pointed at the rest, including the conversation that used to require another person.

There is no secret committee. Attention that can be held, and behavior that can be shaped, already pays for the last two consumer media booms. The second-order claim is simpler than the AGI argument: the companies that win the next decade will be the ones that put an interface on whatever still happens offscreen.

Therapy and Companionship Still Rank First

Marc Zao-Sanders and Sara Biuk spent a year scraping forums, LinkedIn, TikTok and YouTube, then coding 12,637 uses of generative AI from nearly 50,000 records gathered between March 2025 and February 2026. Therapy and companionship was number one, as it had been in their 2025 ranking. They counted more than 1,400 such entries, 11% of the 2026 set, up from 5% the year before.

The study is social listening, not a census, and they say so. The ranking still collides with the industry’s favorite story, which is code, email and search. People do those things. They also take the new medium into the rooms Debord listed as still private.

RANKED USES IN THE 2026 STUDY

Use2026 rankWhat the researchers logged
Therapy / companionship1More than 1,400 entries, 11% of the set, up from 5%
Autonomous agentic operations6New, still small; many cases are notes routed into other tools
Relationship advice7Emotional use, beside love lives, disputes and sex further down
Work buddy8One of 63 top-100 uses that are work-related or overlap with home
Vibe coding21Software written in ordinary language, a new named habit

Users named the bots, assigned them genders, and described a model swap as “identical to losing my friend to cancer.” Others used the same systems to decode a boss’s message or an ex’s text, which is still mediation, just aimed at a human on the other side. Hamilton Morrin, a neuropsychiatry fellow at King’s College London, told the researchers that long waiting lists make the turn to chatbots unsurprising, and that a general-purpose model is not a clinician.

Zao-Sanders flags a second habit he calls thinkslop: handing the first pass of a thought to a model that is built to keep you in the thread. That is a media effect. It does not require the model to be a mind.

What Do Companion Apps Sell?

They sell a person-shaped reply that never needs to sleep, never has another appointment, and never asks for the same care back. Dedicated companion apps sit beside the big chatbots and charge for that reply as a subscription, which is why the category can grow even when a free model will already listen.

By mid-2025 an app-tracking firm, Appfigures, counted about 220 million downloads of companion apps worldwide and $82 million in consumer spend in the first half of the year, on pace for about $120 million for the full year. It logged 337 revenue-generating titles, 128 of them launched in 2025, with the top tenth taking 89% of the money. Those figures are the tracker’s, not a regulator’s, and they still describe a small consumer slice next to cloud bills. They also describe a habit, not a demo.

Replika’s operator has put cumulative registrations above 40 million. Character.AI’s public tallies bounce between monthly actives and lifetime installs depending on who is counting. The honest picture is tens of millions of people who already treat a model as company, plus a much larger group that uses ChatGPT or Gemini the same way without opening a dating-shaped app.

The split in the public argument is not mysterious. One side says a patient listener can be a bridge back to other people. The other says frictionless validation makes the next human conversation feel expensive, and that the people who lean hardest on the product are the ones with the least slack. Both can be true in the same log file. The design incentive is not. A companion that nudges you out the door is a worse engagement metric than one that stays in the chat.

Project Genie, 60 Seconds of a Walkable World

If companions are the native form for talk, Google’s Project Genie is the native form for place. On 29 January 2026, Google Labs opened a prototype to Google AI Ultra subscribers in the United States, aged 18 and over. Product managers Diego Rivas, Elliott Breece and Suz Chambers said the web app lets people create, explore and remix their own worlds from text and images, with Genie 3 drawing the path ahead as you move.

The official limits are the tell. Generations last 60 seconds. Worlds can miss the prompt, lag on character control, and break ordinary physics. Promptable events that DeepMind showed in 2025 are not in the prototype. You can download a video of the walk. You cannot yet live there.

Sundar Pichai, Google’s chief executive, posted the launch himself.

https://x.com/sundarpichai/status/2016979481832067264

In May 2026, at Google I/O, the same prototype gained a Street View pin for some Ultra users in the United States. A real American street can be restyled into an underwater Golden Gate or a black-and-white stockyard. That is the old medium (maps, tourism, the postcard) being used as content for the new one, which is exactly the pattern McLuhan watched with early television and early web mail.

Game makers noticed the obvious threat to pre-built levels. The deeper shift is smaller. For one minute, the generated clip is no longer something you watch. It is somewhere you point a character. Sixty seconds is a toy. It is also a format, and formats get longer.

Hyperscaler Capex Has Quadrupled Since GPT-4

Isabel Juniewicz at Epoch AI, working from SEC filings, added up cash plant spending and new finance leases at Alphabet, Amazon, Meta, Microsoft and Oracle. Combined capital expenditure at those five has been growing at 72% a year since the second quarter of 2023, with a 90% confidence band of 66% to 78%. The series has quadrupled since GPT-4. It was already nearing half a trillion dollars in 2025. If that growth rate holds, Epoch’s line lands at $770 billion in 2026.

THE BUILD EPOCH MEASURED

  • Five filers: Alphabet, Amazon, Meta, Microsoft and Oracle, using cash PP&E plus new finance leases from EDGAR tags.
  • 72% a year: Annualized growth from Q2 2023 through Q4 2025, fitted on log-transformed quarterly totals.
  • 2025 scale: Combined capex nearing half a trillion dollars, after a fourfold rise since GPT-4.
  • 2026 line: About $770 billion if the same growth continues, a naive extrapolation Epoch itself flags as slightly hot versus company talk.

Epoch is careful: capex has quadrupled since GPT-4, and filings do not isolate the AI share. Earnings calls still blame the rise on data centers, chips and power. That plant can train a smarter model. It also runs the companions, the world generators, the briefings and the image pipes that turn leftover hours into sessions. A medium this expensive has to capture time, or the bills make no sense.

Nothing Is Inevitable, McLuhan Wrote

McLuhan’s 1964 line, hosted by his estate as the opening move of Understanding Media, is that the message of a medium is the change of scale, pace or pattern it introduces, independent of the freight. The railway did not invent movement. It invented a different city. Whether the chatbot is “intelligent” is freight. The pattern is that a growing share of comfort, counsel, drafting and play now happens with a system that can run at 2 a.m. without another human waking up.

There is absolutely no inevitability as long as there is a willingness to contemplate what is happening.

Marshall McLuhan, The Medium Is the Massage, 1967

He printed that as a chapter heading, crediting the philosopher Alfred North Whitehead, in a book that came out the same year as Debord’s. The pair do not agree. Debord thought the spectacle chose its machines. McLuhan thought looking hard enough could blunt them. The 2026 usage ranking is what looking looks like: the top job is already company, the native visual form is already a walkable clip, and the capital plan assumes both will need more rooms in the data center.

Contemplation without a change in what gets built is just a better essay. The hatch McLuhan left is still a design choice, a procurement choice and a school choice. It is not a mood.

Sleep, Meals and the Last Unsold Hour

Phone makers are not waiting for world models to finish. They are already shipping on-device AI tools in One UI 9 that turn the lock screen into a briefing and the camera into an editor, another layer between a person and the hour in front of them. The handset is the cheapest way to keep the new medium in a pocket.

WHAT STILL HAPPENS WITHOUT A BOT

  • Unmediated sleep: No companion in the ear, no generated soundscape sold as a night course, no model that notices you woke at 3:14.
  • Meals with people who chew: Food is already logged, photographed and recommended; the remaining gap is eating with someone who is not a feed.
  • Exercise that is not a score: Movement without a coach in the earbuds and without a generated route that wants another lap for the streak.
  • Friendship that can refuse you: The trait companion apps cannot price without breaking the product is another person’s right to be busy, bored or unkind.

Debord’s forecast was that images would start to behave as real beings and pull behavior along with them. A named chatbot that is mourned like a friend is that forecast with a login. A 60-second world you walk instead of watch is the same forecast in another format. Neither requires the model to wake up. Both require a bill, a session, and a person who used to do that hour another way.

Street View inside Genie already lets a paying user pin a real street and restyle it. The park at the end of that street is still there. The open fact is how many people still go, once the generated version is longer than a minute and the companion in their pocket is willing to walk it with them.

Disclaimer: This article is news reporting and analysis of published research, company product notes and public filings. It is informational only and is not medical, mental-health or investment advice, and it is not a recommendation to use or avoid any chatbot, companion app or AI product. Anyone using these tools for distress, therapy or companionship should speak with a licensed clinician, and anyone treating capex or app-revenue figures as a basis for financial decisions should speak with a qualified investment adviser. Rankings, user counts, dollar figures and product limits reflect the cited sources as of 1 September 2026 and can change with the next study, earnings call or software update.

Harry is the editor of NEWS ANALYSIS. He writes across the publication's ten desks, with most of his time going to the stories where a number, a filing or a study decides the argument. His working rule is simple: read the source document before writing about it, and tell the reader plainly which parts are established and which are somebody's claim. He is responsible for the standards set out on this site's Editorial Standards and Fact Checking pages, and for correcting the record openly when the publication gets something wrong.

Continue Reading
Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Trending