There is a new feeling going around, and it is worth naming precisely: the feeling of competence, which is not the same thing as competence itself. AI has made it possible to produce work that carries all the surface markers of expertise, the long documents and structured essays, the strategy papers with their headings and frameworks and confident conclusions, and because the output arrives in seconds and reads well, because it has the shape of something a senior person would write, the person who prompted it begins to feel senior. The tool did not transfer experience. It transferred the appearance of experience, and the gap between those two things is where a whole segment of people now live.

The mechanism is worth understanding. Expertise has always had two jobs, and only one of them is visible. The first is production: you write the report, you build the model, you draft the strategy. The second is judgement: you look at a piece of work and you know, quickly and with reasons, whether it is good or bad. For most of history the two travelled together, because you could not produce serious work without developing the judgement to evaluate it along the way. The apprenticeship was built into the labour itself. AI has severed them. Production is now available to everyone, but judgement is not, and it never was, because judgement is the slow residue of years of doing the work, getting it wrong, watching it fail in contact with reality and adjusting. There is no prompt for that.

So what happens when someone with little experience gets access to unlimited production? The volume itself becomes the evidence. Look how much I delivered, look at this forty page document, look at these five essays this week. The sheer mass of material feels like proof of ability, because in the old world only able people could produce that much, and the signal used to be reliable. Now it is noise. And the moment the output appears on the screen, something psychological happens: the person reads it and thinks, look how brilliant I am, not look how brilliant the model is. The authorship blurs, the prompt was mine so the intelligence must be mine. It is the intellectual equivalent of standing next to a tall building and feeling tall.

The cruel part is that experience is exactly what you need to use these tools well. The more you have done the work yourself, the more useful the model becomes, because a seasoned operator reads an AI draft and reacts in seconds: this section is good, this claim is wrong, this framework is generic and would collapse in our context, this paragraph sounds impressive and says nothing. That reaction is the whole game. The model produces, the expert filters, and the combination is genuinely powerful. But if you cannot filter, you cannot tell the difference between output that is good and output that merely looks good, and everything the model produces seems equally polished because polish is the one thing it never fails at. So you ship it all. You forward the document, you publish the essay, you present the strategy, and to another second hand thinker it lands, because it looks like the real thing. The room fills with people trading documents none of them can properly evaluate, each one reinforcing the other's sense of competence, and none of it makes sense, and nobody in the loop can tell.

The name fits for another reason. Second hand smoke harms the people nearby more than the smoker realises, and the same is true here. The person generating unfiltered output feels fine; it is everyone around them who breathes it in. The colleague who wastes a day reviewing a document with no thinking inside it, the team that builds on a strategy nobody actually reasoned through, the decision made downstream on foundations that were never inspected. The damage does not land on the producer. It drifts into the room.

This is why I have stopped being impressed by volume. A long document tells me nothing now. What tells me something is the conversation after the document: ask the person to defend a claim on page twelve, ask them what they would cut, ask them what the model got wrong. The ones with real expertise light up, because editing is where they live. The second hand thinkers go quiet, or worse, they generate another document. There used to be a shortcut to looking experienced, but there has never been a shortcut to being experienced, and AI has widened that gap into a canyon while filling the canyon with fog. From a distance, both sides look the same.

The honest use of these tools is humbler than how people using them suggests. f you are early in your career, use AI to learn faster rather than to perform seniority you have not earned: interrogate the output, find its errors, treat every draft as a sparring partner rather than a finished product with your name on it, because that is how the tool builds judgement instead of replacing the need for it. And if you are experienced, this is your moment. Your judgement was always the scarce asset, and it just became scarcer, because production is free and evaluation is not. The market will take a while to figure this out, and it will be fooled by volume for a few more years, but then the documents will start failing in contact with reality, as documents do, and it will become very clear who was thinking and who was forwarding.

First hand thinking was always hard. It still is. The only thing that changed is how easy it became to fake.

‍