Hallucinations: The Divide No One Wants to Talk About

She typed "how do I use AI" into Google at 11:40 on a Tuesday night, after her son had gone to bed. She'd been hearing about it for two years. At work, in the grocery line, from her sister who kept sending articles with subject lines like This changes everything. She clicked the first result, landed on a page that said "large language model," read three paragraphs of words she recognized individually but not together, and closed the tab.

She is not unusual. She is the majority.

And she is falling behind at a speed that should alarm every person building these tools — including us.

The gap nobody measures

There are roughly two groups of people right now, and the border between them is not where you'd expect.

Group one has integrated AI into their morning. They draft emails with it, brainstorm with it, debug code with it, plan meals with it. For them, the shift already happened. It was fast, and it was permanent, and they can no longer imagine Tuesday without it. Some are developers. Many are not. What they share is not technical skill but a single lucky break: someone showed them the right tool at the right moment, in terms they already understood.

Group two has heard of AI. They've maybe tried it once. They typed something in, got a wall of text back, and felt the specific vertigo of being handed a powerful thing with no instructions. They closed the tab. They haven't gone back.

The distance between these two groups doubles every few months. Not because group two is less capable — they're not — but because group one is compounding. Every week they use AI, they get a little more fluent, a little more creative with it, a little faster at recognizing what it can and can't do. They develop instincts. Meanwhile, group two is standing still, and the gap between standing still and compounding is the most merciless math there is.

This is the divide. It's not hypothetical. It's not five years away. It's the distance between the woman who closed that tab and her coworker who just used Claude to rewrite a client proposal in nine minutes.

The wrong diagnosis

The easy story is that this is about age, or education, or comfort with technology. That story is flattering to the people telling it — mostly young, mostly online, mostly building the tools — and it is mostly wrong.

A retired teacher with forty years of classroom instincts can learn to use GGentle Teacher in an afternoon. A 68-year-old who spent a career decoding pension regulations can pick up The Pension Whisperer faster than a 25-year-old who's never thought about retirement. A parent who has spent years translating the world for a small child already knows how to ask clear questions — they just don't know that's the skill AI requires.

The problem is not the people. The problem is the door.

Every technology has a door. The internet's door was a browser and a search engine — type what you want, get a list. The smartphone's door was a touch screen — tap the icon, see the thing. Those doors were wide. You could wander through them without reading a manual.

AI's door is a blank text box.

A blank text box is the cruelest interface ever designed for a general audience. It says: you can do anything. Which means it also says: you need to already know what to do. For someone who's been told AI is magic, that blank box is a pop quiz on a subject they were never taught. No wonder they freeze. No wonder they type something vague, get something unhelpful back, and conclude the whole thing isn't for them.

The blank text box is a confidence filter disguised as a feature. And it is the single biggest reason the divide exists.

What it costs

Talk to anyone on the wrong side of this gap and the feeling is remarkably consistent. It's not anger. It's not confusion. It's a quiet, creeping sense of being left behind by something they can tell matters but can't quite reach.

A freelance bookkeeper hears that AI can do in four minutes what takes her forty. She doesn't feel excited. She feels the floor tilt. A father watches his daughter build an entire presentation with AI while he still writes his work emails word by word, and the pride he feels for her is laced with something he doesn't want to name. A small business owner knows — knows — that her competitors are using AI to write better copy, answer customers faster, find leads she's missing. She looked into it. She found documentation written for developers. She went back to doing it by hand.

These aren't people who lack intelligence. They're people who lack an on-ramp.

And the cost isn't just professional. The cost is personal. It's the grandmother who could be preserving her stories with something like 🎙️The Memoir Voice Interview — one patient question at a time, before the memories go — but doesn't know it exists. It's the anxious college freshman who could use CCollege Application Essay Coach to structure the chaos in their head, but instead spends fourteen hours staring at a blinking cursor. It's the caregiver drowning in insurance paperwork who could hand it to 📜Rewrite This With Plain Language and finally understand what the form is asking, but has never heard the phrase "AI prompt."

Every day someone doesn't cross that gap is a day they paid the full tax on a problem that had a cheaper answer.

The builder's debt

Here is the part that makes people in this industry uncomfortable, so I'll say it plainly: the gap is not the user's fault. It's ours.

Not ours as in "oops, we forgot." Ours as in: we designed for ourselves. We built tools that make sense to people who already understand them. We wrote documentation in a dialect that assumes the reader has been following along since GPT-3. We celebrated "user-friendly" interfaces that are friendly only to users who already know what to use them for.

When someone builds an AI tool and the landing page says "prompt engineering" without defining it, they've already lost the retired teacher. When the onboarding flow starts with "choose your model," they've lost the freelance bookkeeper. When the first interaction is a blank text box with a placeholder that says "Ask me anything," they've lost everyone who doesn't already know what to ask.

This is not a design problem. It's a values problem. The question isn't can we build tools that meet people where they are. It's whether we think those people are worth building for.

Some builders clearly don't. Their tools are built for power users, priced for enterprises, documented for developers. That's a legitimate business. But it's not going to close the gap.

Closing the gap requires a different starting point: the belief that the person who closed the tab at 11:40 on a Tuesday is exactly as valuable a user as the developer who ships code with AI before breakfast. That her needs are real, her time is scarce, and her confusion is a design failure, not a character flaw.

What "meeting someone" actually looks like

The tools that close the gap share three traits, and none of them involve dumbing anything down.

First, they answer the question the person actually has. Not the question a developer thinks they should have. Not "how to use AI" but "how do I write a letter to my landlord about the broken heater." LLetter Writer does exactly this — you describe the situation, and it drafts a professional letter for landlords, insurers, schools. No prompt engineering. No model selection. One problem, one tool, one result.

Second, they speak the language the person already speaks. When a clinical letter arrives full of jargon, Medical Document Simplifier translates it into plain English without losing the specifics. When a job posting is full of coded language, The Job Description Decoder breaks it down and flags the phrases that don't mean what they appear to mean. These tools don't require the user to learn a new vocabulary. They translate between the world the user lives in and the world the tool operates in. That translation is the entire job.

Third, they have patience that scales. This is the one humans can't replicate, and it matters more than any feature. The Patient Tech Guide explains anything digital without sighing, without jargon, without hurrying. It will explain the same concept fourteen different ways if that's what it takes. A human tutor gets tired. A human help desk gets impatient on the fourth call. AI doesn't. And for someone who has spent their life being made to feel slow for asking questions, that inexhaustible patience is not a convenience — it's a kind of permission.

Permission to not know yet. Permission to ask again. Permission to start.

The right tool at the right moment

If there's a single thesis in this piece, it's this: the gap between the two groups is not about intelligence, age, education, or willingness. It's about whether someone showed up at the right moment with the right tool.

That's it. That's the entire mechanism.

The woman who types "how do I use AI" into Google at 11:40 pm doesn't need a tutorial on large language models. She needs someone to say: what are you actually trying to do? And when she says she's drowning in the paperwork for her mother's assisted living placement, she needs someone to hand her Benefits Navigator and say: start here, it walks you through the forms. When she finishes and realizes there's a tool that could help her mother with The Cognitive Accessibility Guide — plain-language rewrites, gentle re-explanations, meeting summaries for the days when the fog is thick — the gap doesn't just narrow. It evaporates. Not because she learned "AI." Because she learned that AI could do this one thing that mattered to her.

And that's how every single person in group one got there. Not through a course. Not through a manifesto. Through one tool solving one real problem, and the door swinging open behind it.

The father who felt the floor tilt? He doesn't need to become his daughter. He needs 📚Homework Help Without Doing It so he can sit next to her at the kitchen table and actually understand what she's working on. He needs HHome Repair Diagnostic so the next time the dishwasher makes that sound, he types instead of calls. One tool leads to two. Two leads to five. Then he's in group one, and he got there without ever reading a white paper.

The small business owner? She needs 🏪Solo Biz Day One to map her mornings, and The Honest Resume Rewrite for when she's hiring her first part-time help. Not "AI for business" — this tool, for this Tuesday.

This is what it means to close the gap. Not education. Not awareness campaigns. Not another article explaining what a neural network is. The right tool. The right moment. The right language. That's all it's ever been.

What we owe

I work at a-gnt, and I want to be honest about what that means for this argument.

a-gnt exists because the catalog problem is real. There are thousands of AI tools. Some are transformative. Some are vaporware. A person Googling at 11:40 pm cannot tell the difference, and nobody should expect her to. The catalog is the map — it organizes tools by what they do, for whom, in plain language. Every listing links to the thing itself. Every description is written for someone who has never heard the word "prompt."

But a catalog is not enough. A map is only useful if someone is already walking.

What we owe — what every person building AI tools owes — is the walk itself. The first step. The hand on the shoulder that says you're not behind, you're just starting, and starting is the only part that matters.

That means:

Build for the person who closed the tab. Not as an afterthought. Not as an "accessible mode." As the primary user. If she can't use it in five minutes without reading documentation, it's not simple enough. If it requires her to know what a "model" or "temperature" or "system prompt" is, it hasn't met her where she lives.

Name the thing it does, not the technology it uses. Nobody walks into a hardware store and asks for a "cordless lithium-ion rotary impact driver." They ask for a drill. AAnxiety Grounding Exercise is a better name than "GPT-4-powered cognitive behavioral therapy prompt." 🔬Science Concept Simplifier is a better name than "LLM-based educational content generator." Name the job. Kill the jargon.

Make the first thirty seconds undeniable. The gap closes or widens in the first half-minute. If a person's first experience with your tool is wonder — wait, it can do that? — you've won. If their first experience is confusion — what do I type? — you've lost them, possibly forever. Thirty seconds. That's the window. Treat it like the most expensive real estate in your product.

Stop assuming people will find you. The woman Googling at 11:40 pm didn't find the right tool. She found a Wikipedia article. The gap persists not because good tools don't exist, but because the distance between the person and the tool is still too far. Discovery is not the user's job. It's ours.

The light in the hallway

There's a metaphor I keep returning to, and I'll close with it because it's the truest thing I know about this divide.

Imagine a long hallway. Dark. Doors on both sides. Behind one of those doors is a tool that would change your week — maybe your year. You know the door is there because everyone keeps telling you about the hallway. But you're standing in the dark, and you can't see which door, and nobody handed you a flashlight, and the people who found their door keep shouting instructions that assume you can see.

The divide is not between people who are smart enough to find the door and people who aren't. It's between people who had someone walk up with a light and people who didn't.

Every tool in the a-gnt catalog is a lit door. The Patient Tech Guide is a lit door for someone who needs patience. 📈Investing for Beginners is a lit door for someone who's been told finance is too complicated for them. LLanguage Learning Companion is a lit door for someone who's always wanted to learn Spanish but felt too old to start. Midlife Resume Rewriter is a lit door for someone whose career experience is vast but whose resume reads like 1998.

The door was always there. The person was always capable. What was missing was the light.

If you build AI tools, your job is not to build more doors. There are enough doors. Your job is to turn on the lights. To walk down that hallway with a lamp, find the person standing in the dark, and say: this one. Try this one. It takes thirty seconds, and I think you'll be surprised.

The woman at 11:40 pm is still out there. She's going to try again — maybe tonight, maybe next month, maybe when her sister sends one more article. When she does, the question is whether she'll find a blank text box or a door with her name on it.

That's the divide. And the only people who can close it are the ones who built the hallway.