You have tested the free version. That is a crucial difference.
The car park manoeuvre
Imagine someone gets into a car for the first time. Not just any car — a modern electric vehicle with adaptive cruise control, autonomous lane changing and a powertrain delivering 500 horsepower. But the person never starts the engine. He sits in the driver's seat, turns the wheel, notes that the seat is comfortable, and then gets out.
"I have tested an electric car," he says. "It didn't move. Overhyped."
That is roughly where the public AI debate is right now.
What most people have actually tested
When a journalist, commentator or politician says they have "tested AI" they almost always mean that they have opened a free chatbot in a web browser, asked a question or two, and judged the result. That is not unreasonable — that is how we test most digital services. But AI is not most digital services.
What they have actually tested is the lowest tier of a technology that spans an enormous capacity spectrum. It is like evaluating modern healthcare by visiting a pharmacy. The pharmacy is not bad — it does its job — but it is not healthcare. It is the entry point to healthcare.
The free version of an AI service is the entry point. It uses a simpler model, with shorter reasoning time, without access to tools, with limited context window, and without the ability to perform multi-step tasks. What it delivers is fast, shallow answers. That is what it is designed for.
The spectrum nobody talks about
Public debate treats "AI" as a single thing. But the difference between a free model and a top model with full tools is not a difference of degree — it is a qualitative difference. It is the difference between a calculator and a computer.
Let me be concrete without making this a product review.
Reasoning time. A free model answers instantly. That means it does not think — it reacts. The most capable models have the ability to reason before answering, sometimes for tens of seconds. That sounds like a trivial difference. It is not. Think about how you yourself function. If someone asks you a question and demands an answer within three seconds — what do you do? You guess. You say the first thing that occurs to you, regardless of whether it is correct. Especially if there are no consequences for being wrong. But if the same person says "take a few days, this is important" — then you seek information, weigh alternatives, and deliver a considered answer. AI functions in exactly the same way. The free model guesses. The full model thinks.
Context window. Free models handle short texts. Paid models can work with hundreds of pages simultaneously — entire reports, code projects, book manuscripts. That means they can see patterns, contradictions and connections that a human would need days to identify.
Tools. A free model can only write text. A full model can search the web, run code, create and edit documents, build presentations, analyse data, generate images and link to external systems. It is the difference between having a conversation partner and having a conversation partner with an office, a workshop and a library.
Model selection. Free versions run the cheapest model. Paid versions give access to the strongest models — ones that perform at expert level in medicine, law, physics and programming on standardised tests. That is not marketing. That is measurable performance on the same tests used to certify doctors and lawyers.
An invisible chasm
Here a problem emerges that is rarely discussed: the person who has tested the free version does not know what they don't know.
If you have never seen an AI reason through a complex problem step by step, build a functioning tool from a half-formed idea, or hold together a project over dozens of iterations — then you have no frame of reference for what the technology can actually do. You think you do, because you have "tested". But you have tested a stripped-down version designed to be free, fast and broad — not deep.
That creates an invisible chasm. Those who use AI's full capacity daily — researchers, engineers, developers, consultants, writers — live in a different reality from those who base their opinions on the free version. And those who live in that other reality struggle to explain the difference, because it must be experienced to be understood.
Consequences for the debate
This is not an academic question. It has direct consequences.
Regulation. If legislators base their understanding of AI on the free version — which they very likely do — they will regulate a technology they do not understand. They will either underestimate the risks (because the free version seems harmless) or overestimate the shallowness (because the free version seems stupid). Both misassessments lead to bad policy.
Education. If teachers and school leaders judge AI's capacity based on the free version, they will make the wrong decisions about how AI should be integrated — or not integrated — into teaching. They will miss that the tool already today can function as a high-quality personal tutor, but only in its full version.
Business. If business leaders dismiss AI as "chatbot hype" after testing the free version, they risk missing the most significant productivity improvement since computerisation. Their competitors — those who actually invest in understanding the technology — will not make the same mistake.
Opinion-forming. If journalists write about AI based on the free version, they produce misinformation. Not intentionally, but effectively. They inform the public about a technology they themselves have not understood, and the public makes decisions based on the false picture.
"But surely it should work anyway?"
A common objection: if the technology is so good, shouldn't even the free version be impressive?
The answer is: it is impressive — for what it is. A free model in 2026 is more powerful than the best available AI was two years ago. But we don't judge technology against what it was — we judge it against what it claims to be. And what AI sceptics react to is the gap between the hype and their own experience. That gap is real. But the reason is not that the technology is overhyped. The reason is that they are testing the wrong version.
It is like complaining that Netflix's streaming quality is poor — when you are watching via a free VPN over a 3G connection. The problem is not Netflix.
The democratic dilemma
There is a valid objection in all of this: if AI's full capacity is only available to those who pay, it creates inequality. Those who can afford a subscription — often already privileged — get access to a tool that makes them even more productive. Those who cannot, fall behind.
That is true. And it is a problem that needs to be addressed. But the solution is not to pretend the gap does not exist. The solution is not to let uninformed statements about "chatbots" shape policy. The solution is to acknowledge the capacity difference, make it visible, and then discuss how we ensure broad access.
It starts with honesty about what the technology can actually do.
The appeal
If you have opinions about AI — and you should, it concerns us all — make sure your opinions are based on the right data. Don't test the demo version and draw conclusions about the whole thing. That is like reading the back of a book and reviewing it.
Invest a month with a paid version. Give it a real project — not a trivial question, but a problem you actually grapple with in your work. Observe the difference. Then form your opinion.
If your opinion after that is still that AI is overhyped, at least you have an informed opinion. That is all I ask.
But I suspect your opinion will change.