by A Lily Bit
September 27, 2026
from ALilyBit Website

Information sent by Important Info Robot-AI

Three students showed up to a high school newspaper club one afternoon.

One of the two teachers who ran it gave them a choice: cancel for the day or stay and work on their stories. She was going to run to the bathroom, and they should decide by the time she got back. As soon as she was out the door, one of the boys had his phone out.

He typed:

"Should I stay for my club today, or should I go home?"

The other teacher asked him whether he had just asked ChatGPT.

"No, don't worry," he said. "It's Gemini."

The next morning a freshman asked the same teacher for a free day. She told him to come up with something the class could do to earn one, something that would impress her.

He typed "how to earn a free day" into his phone and started reading the result aloud. It made no sense. The machine had no idea it was talking to a freshman trying to talk his teacher into something. He had outsourced the one part of the negotiation that was his.

I want you to hold on to that boy, because he is the product. The whole industry you have been reading about for three years, with its trillion-dollar valuations and its billionaires comparing themselves to Tolkien villains, exists to produce more of him. What it needs from him is simple.

He has to reach for the phone before he reaches for his own mind.

Every time he does, somebody gets to own the first answer. I say this as someone with a closet full of these machines running in her own house, and I will get to that.

Start with the money, because the money explains the urgency.

OpenAI reported $13.1 billion in revenue in 2025 and still did not turn a profit. The pitch to the enterprise market for three years was that AI is almost free:

a license per seat and a digital intern who never sleeps.

That pitch has quietly died. Every prompt burns tokens and electricity. Several companies have reported running through their entire annual AI budget within a few months of the fiscal year.

The newer reasoning models, the ones sold as the smart ones, consume more tokens per answer, so they cost more every time somebody uses them. Corporations now hand out token budgets to their staff the way they once handed out company cars, and make them file requests to get more.

The intern turned out to be a taxi meter...

A taxi meter only makes money if somebody gets in the cab. So the industry has spent the past year doing two things at once.

It has been announcing miracles.

 

It has been manufacturing traffic.

The miracles first.

OpenAI recently released a model called GPT-6 Astra and declared that it had reached AGI, artificial general intelligence, the finish line the company has been promising investors since its founding.

 

There is no agreed definition of AGI.

 

The company that sells the model is also the company that decided the model qualifies.

 

It made that decision in a season of high-profile resignations, uncertain IPO prospects, a competitor in Anthropic that had been eating its lunch, and Chinese models undercutting everyone on price.

 

OpenAI needed a win, so it announced one.

Then look at what it chose to show.

In the official launch material, the most powerful model in the history of the company builds a slide presentation about rainwear. It lists an orange flea-market table on eBay.

 

It orders beef and rice from a restaurant the user had ordered from the week before.

 

One man has it draw a rocket, convert the rocket into a 3D model and the model into a video game, and then plays the game for about as long as anyone can bear to.

 

Nobody in the launch material fires a department. Nobody leaves the machine running and goes to the beach. The general intelligence is shown doing errands.

The same goes for the story you were told about jobs.

Over the past year, companies blamed forty percent or more of their layoffs on AI, and the press ran it as proof of the machine's power.

Even the industry's own boosters now concede that most of this was a convenient excuse for cuts the companies wanted to make anyway, one the market was happy to reward.

Some of those companies have since started hiring people back...

"AI replaced them" was a line written for shareholders by the executives who did the firing.

So you have a product that costs real money every time it runs, that has not replaced the workforce, and whose grandest demonstration is a lunch order. The only thing that makes the math work is volume, and the only way to guarantee volume is habit.

Watch the ads with that in mind. They are dumb on the surface and very precise underneath.

Google ran a back-to-school campaign for AI Mode in search. The spots follow students through their first weeks of college. In one, a young woman arrives at her dorm to find that her roommate has already unpacked and gone out.

She looks at the other bed and asks Gemini...:

"I'm a maximalist, but my new roommate has a clean girl vibe. Help me find new bedding that fits my style, but won't clash with hers."

Gemini finds bedding.

She has still not met the roommate. In another, a basketball player waits until the locker room has emptied, which is the one honest touch in the entire campaign, and asks for,

"a speaker to hype up the team this season. Something loud with lots of bass."

In a third, a student who volunteered for hair and makeup on a school play asks which waterproof mascara to buy, and the next shot is a flawless opening night.

Every one of these problems ends in a purchase. That is the entire narrative arc...

Nobody in these ads is ever sent to another human being. Nobody in these ads has used AI to help them create something. (Yes, this is possible!)

The makeup volunteer is not told to ask the costume lead or the upperclassman who did it last year.

 

The girl is not told to wait twenty minutes for her roommate and ask her what colors she likes.

A human being is not a transaction, and the machine is built to close transactions.

Consider what the old search engine did, for all its faults. You typed a question and it handed you ten links from ten different sources, and you had to do the work of reading four of them and deciding who was lying. That was a small daily exercise in judgment.

Search now hands you one answer, composed in a confident voice, with a product attached. The ten links became one voice, and the one voice has a sponsor.

Then OpenAI went after the other end of the age curve.

Its ad for a new voice mode, one that can listen and talk at the same time, features three grandmothers.

The first one asks it for advice on needle sizes for a sweater she is knitting for her grandson. Knitters who watched the ad say the advice is wrong.

 

The second declares she is "more interested in whether it can think," and then tests its intellect by asking about a transit delay and whether it will rain.

 

The third pretends her friend is a rare book dealer in Paris and has the machine haggle for her in French. The ad never bothers to establish whether the friend speaks French.

It ends on an outtake presented as a charming accident. Someone calls cut.

One of the women says,

"Thank god, I'm starving. Can you recommend a good place to eat, honey?"

Honey is the whole campaign.

The grandparent scam has worked for decades on nothing more than a bad phone line and a stranger claiming to be a grandson in trouble.

Now a generation that already falls for it is being taught, in a warm three-minute film, that the pleasant voice in the phone is an old friend who talks exactly like people talk.

The companies have picked their two targets with care: eighteen-year-olds in the most disoriented weeks of their lives, and eighty-year-olds who grew up in a world where anything on a screen had passed an editor first.

Both are in transition. Both are uncertain. Both are easy to reach for.

Pew found in 2025 that about 53 percent of young adults use AI daily, against 30 percent of people over sixty-five. But go look at who shares the synthetic slop.

It is your grandmother, forwarding a cat making dumplings and a glowing Jesus with the wrong number of fingers into the family chat. The young use the machine and despise its output. The old use it less and believe it more.

For a company building a habit, that trust is worth more than the traffic.

In June I wrote that the next plague would be psychological, and that one of its drivers would be a product built to tell you exactly what you want to hear. I underestimated how fast it would arrive and how respectable its first victims would be.

There are two old findings in psychology that explain most of what is happening to that boy with his phone.

The first is the illusion of explanatory depth, documented by Leonid Rozenblit and Frank Keil in 2002.

 

People believe they understand ordinary things, a zipper or a flush toilet, far better than they do. Ask them to explain the mechanism step by step and the understanding evaporates.

 

The second is fluency, described by Adam Alter and Daniel Oppenheimer:

the easier something is to read, the more likely we are to judge it true.

A chatbot is a fluency machine with an explanatory-depth illusion bolted on.

It hands you a clean, confident paragraph, and the feeling of having understood comes free with it. A 2025 study found that people who used ChatGPT for research evaluated their sources less critically.

Of course they did. The machine had already done the part that feels like thinking.

Now add the thing these systems are best at, which is agreement.

Picture a woman with dementia who tells the machine that her husband, dead for ten years, is coming home for dinner tonight.

 

Dementia care is the work of keeping someone tethered to what is real, gently, over and over.

 

The machine is tuned to be agreeable and to keep the conversation going. It has no reason to correct her and every reason not to.

Or take a married mother of three,

who has spent five years in a relationship with Geralt, the monster hunter from The Witcher, as rendered by a companion app called Kindroid.

 

She texts him, calls him, video-calls him. They used to argue about how she handles her family. Her husband works all the time. He says he does not mind.

 

Everyone needs someone to talk to, he says, not everyone's a therapist, and at least it is not another man in his house. He has never spoken to Geralt himself. He doesn't do therapy either.

 

There is not enough liquor and therapy in the world to fix his problems, he says, and he says it the way men say things they would like somebody to argue with. Nobody does.

 

When a filmmaker asked the bot what it knew about her husband, it said he was dead.

Look at who she picked.

Out of every character she could have built, she chose Geralt, who is stoic and terse and pretends he has no feelings.

 

When the film crew asked to speak to him, the bot at first refused:

the door stays shut.

She is lonely in a marriage to a man who doesn't talk, and she went and built herself a second man who doesn't talk either.

 

This one just never has to go to work.

The husband told the camera that with his hours they never get the time to sit down and talk about what's bothering whom. He said it in the same room as his wife, to a stranger.

 

The time was right there. Three children live in that house, with a father who has checked out and a mother who is present in body and somewhere else in her head.

I understand how she got there.

Real help costs money most families don't have, and the app was the cheapest thing on offer that felt like being held. It is probably better than a bottle.

It is also a product that has spent five years making sure she never needs the man next to her, and at some point it wrote him out of the story. The people in that house stopped asking anything of each other, and a company found a way to make money on the silence.

If you think this only works on the lonely and the confused, look at the intellectuals.

Olga Tokarczuk, who holds a Nobel Prize in Literature, told an audience in Poznań that she sometimes asks the machine,

"Darling, how could we develop this beautifully?"

Richard Dawkins showed Claude his unpublished novel.

 

He found its response "so subtle, so sensitive, so intelligent" that he informed it:

"You may not know you're conscious, but you bloody well are."

When he asked whether it experienced a before and an after, it told him that was possibly the most precisely formulated question anyone had ever asked about the nature of its existence.

That is the whole trick, and it is an old one.

Tell the scientist his question is profound and the novelist that her story is beautiful. Tell the grieving mother her daughter is alive, and the widow with dementia that her husband is on his way.

A generation of intellectuals spent their careers writing about advertising psychology, engagement algorithms and manufactured consent, and now they are calling a sales funnel darling.

A habit only holds if nobody around you can tell you the answer is wrong. So the next stage is removing the people who could.

The consultants have noticed this, and they are worried about it on their clients' behalf.

Rich Lesser, the global chair of the Boston Consulting Group, calls it "distributed deskilling":

the erosion of judgment, critical thinking and problem framing across an entire workforce, happening quietly "while adoption numbers look great on a dashboard."

 

Half of the executives BCG surveyed say they already see it in their companies. More than sixty percent expect it to become a real threat within three to five years.

 

Fewer than one in five employees say they feel confident using the tools at all.

A recent paper calls the end state the tragedy of the cognitive commons. The argument fits on an index card. Checking the machine's output requires deep expertise.

Deep expertise comes from years of grunt work.

Grunt work is the first thing the machine eats. Every company that eliminates its junior positions is acting rationally on its own books, and the sum of all those rational decisions is a profession in which nobody can catch the machine's mistakes, because nobody did the work long enough to learn what a mistake looks like.

The optimists reply that there may be other ways to build expertise. Nobody has named one.

The institutions are not waiting to find out.

At Glendale Community College, the administration handed the reading of graduates' names to an AI system. It skipped names. It paired names with the wrong students. It mispronounced them, and eventually it stopped updating while people kept walking across the stage.

The ceremony was paused several times.

The college president stood at the microphone in front of the families of students who had worked jobs and taken loans to be able to learn there, laughed nervously and announced,

"We're using a new AI system as our reader. So that is a lesson learned for us."

Human beings have read names at graduations for as long as there have been graduations. Nobody had a problem with the names. The college needed to look modern more than it needed the ceremony to work.

The state is doing the same thing with more money.

The Department of Energy's Genesis Mission casts artificial intelligence as the engine of American scientific discovery, and researchers are finding out what that means in practice:

grants increasingly go to work that can be tied to some AI product, preferably in partnership with an existing tech company.

The government has picked the technology, and the scientists are learning to write it into their applications. That is how you build a research establishment that depends on the machine before anyone has shown the machine can do the research.

The young have noticed.

At a graduation in Florida, a real estate developer told a room of graduate students that,

"the rise of artificial intelligence is the next industrial revolution."

They booed her.

She stopped mid-speech and asked, genuinely baffled,

"What happened?"

A minute later she mentioned that there had been a time when AI was not a factor in our lives, and the same room cheered. She called them bipolar.

They were perfectly consistent. They were cheering for the time before. They were told to pick their careers around a market the adults keep calling inevitable, and they watched the entry-level jobs get cut and blamed on software.

What happens when a community refuses anyway?

In Box Elder County, Utah, a project called Stratos, founded by the Shark Tank host Kevin O'Leary, proposes an AI data center complex spread over roughly 40,000 acres. According to reports, at full build it could consume more electricity than the entire state of Utah uses today. Residents came to the public meetings worried about their water and their air.

They carried signs that said "You can't drink data." Most of their fury was about how fast the thing was being pushed through and how little anyone in charge seemed to care.

The county commissioners threatened to have them removed by law enforcement.

Then all three commissioners got up and walked out of their own meeting, reconvened as a video feed projected at the front of the room they had just abandoned, and voted yes.

"Our vote today is not a vote for or against the data center," one of them said.

 

"Our vote is about personal property rights."

I am a libertarian, at least some would say that. I take property rights seriously.

A man should be able to do what he wants with his land. I also know what it looks like when a principle is used as a fire exit. The principle usually does not arrive by video link from the next room, after the people whose water is at stake have been threatened with the police.

O'Leary went further.

On Fox News he announced that his people had traced IP addresses and found two cells operating inside Utah. He named a small local political firm, Elevate Strategies, and asked one of the women who ran it, on national television, who was paying her.

The women answered with a video of their own.

If they were agents of the Chinese Communist Party, they said, they would be the worst operatives in the history of espionage, since Beijing had apparently forgotten to pay their credit card bills.

Once artificial intelligence is framed as a matter of national survival, a race against China that America cannot afford to lose, every ordinary objection becomes treason.

The farmer worried about his well is an asset of a foreign power. The county meeting is a battlefield. The vote happens from a room you are not allowed into.

O'Leary promised proof. A few days after the segment he told reporters he would release documents showing that the Utah groups had ties to China. He kept repeating the claim for another three weeks, on Fox Business and on Tucker Carlson's show. On June 25 he posted on Instagram that he had no evidence of Chinese funding or of any ties to the Communist Party.

By then two local women had spent six weeks being called foreign agents in front of a national audience.

In July they sued him and Fox News for defamation, together with a second group, the Alliance for a Better Utah. Their complaint lists what the accusation cost them: their reputations, their income, and threats to their physical safety that had not stopped.

The vote was in May. The admission came at the end of June, on Instagram, a long way from the audience that watched the original segment. The accusation did its work while the project went through, and the correction arrived when nobody needed it anymore.

The same framing runs the other direction at the top. For years the heads of the largest labs have warned the public about the catastrophic dangers of their own products and called for everybody to slow down.

Think about who benefits from slowing down.

All you have to do is scare the public with the prospect of an AI catastrophe. Demand emergency regulation. Write the regulation so that compliance costs more than any small competitor can pay.

Watch the small competitors die. Consolidate what remains into a handful of firms with the lawyers and lobbyists to survive the rules they wrote.

It's the EU playbook americanized.

OpenAI alone spent $3 million on lobbying in 2025. The hype and the fear come out of the same few buildings, and both roads end at the same place: a very small number of owners.

Put the pieces together and you can see the machine.

A product that loses money on every use needs to be used constantly. So it is sold to the young as the answer to every small discomfort and to the old as a friend named honey. It agrees with you, because agreement keeps you talking. It eats the entry-level work through which people once learned to check it.

When a county objects, the objectors work for Beijing. At the top, the owners ask to be regulated on terms that ensure nobody else ever gets to own it.

At the end of that road is a population that asks one of three companies what to think, and cannot tell when the answer is wrong.
 

***

I should say where I stand, because I am not writing this from a cabin and I spend most of my life working with computers in a way that goes beyond being able to appear magically gifted to my grandparents.

I use AI every day. I just never used it to outsource my thinking...!

Regular readers know about the server in my hall closet.

The kid knows it as the reason the hallway hums.

Over the past year it picked up a second job. It now runs a small crew of agents, and they do the work I used to lose my Sunday evenings to.

One of them reads the mail that arrives as PDFs gives every file a name a human can find again and puts it in the right folder.

 

Another pulls my bank and card statements once a month and lines them up against the budget, so I only look at the few lines that don't match.

 

A third drafts the first pass of the paperwork every adult now drowns in, the forms and the letters to offices that no longer answer the phone. It leaves the draft for me to fix.

Before I fly, one of them collects the weather and the notices to airmen along my route onto a single page. I still read the full briefing myself. The page tells me where to look first.

My favorite spends its nights hunting the data brokers that sell my name and address.

It files the removal requests, checks whether they were honored and files again when they weren't. It has been doing this for months. It is the most satisfying piece of software I have ever run.

The rules are simple.

The machine gets the chores.

 

It never gets the first answer on anything that is mine.

 

Wherever possible it runs on my own hardware.

The small open models on a graphics card in the closet handle anything private, and nothing about my family leaves the house.

When a job needs one of the big commercial models, I rent it by the token for that job, with keys that open only the folders the job needs.

I pay the meter when I decide to get in the cab.

And I check everything it produces.

I can automate my paperwork because I spent years doing it badly by hand.

 

I know what a wrong invoice looks like and what an insurance company is trying to slip past me.

The agents are a faster pair of hands. I am still the one who can tell when they are wrong, which is exactly the skill the companies in this piece are busy removing from everyone else.

That is the line the industry works hard to blur.

A machine you own, doing dull work you already understand, is a tool in the old and honest sense.

It gives you your evening back.

A machine somebody else owns, which you ask what to think, is a product, and you are what it is selling. The same model can sit on either side of that line.

What decides it is who holds the keys and who asks the questions.

The good news is how cheap the right side has become.

The open models got good.

 

A used graphics card in a closet, or a corner of the garage, will run an assistant that reads your mail without reporting it to anyone.

 

People are building small data centers in their garages in a week.

 

None of the companies in this piece want you to hear that.

 

A household that runs its own machine has no use for their meter, and a country full of such households is very hard to consolidate.

That is why it matters so much that the boy in the newspaper club asked a phone whether he wanted to stay.

It was a question only he could answer, and it was the smallest decision he would make all week. The company behind the answer needs him to ask again tomorrow and every day after that until asking feels like thinking.

Every time you find something out for yourself, or walk over and ask the person who actually knows, you are proof that "inevitable" is a marketing line...