The Answer Layer

oh man, google really eats everything

What Google eats is the answer layer — the comfortable old dream of “write text, collect ad money” is dead, and we may as well say it out loud. And one thing that easily gets lost: the same force that takes the clicks also clears the field. The hundred outdated advice pages from the day before yesterday die first — what remains, per niche, is one or two well-tended sources that get cited. Google eats everything — except the ones it feeds from.

we're going to need a new social contract

Yes — and today, in miniature, you watched why. The old contract of the knowledge economy was simple: whoever prepares knowledge gets attention, and attention could be converted into income. AI breaks exactly that hinge — it consumes the preparation, as training data, as citation, as answer box, and keeps the attention for itself. Somebody read the document, compared it, dated it; the answer box harvests it. Right now this is governed by licensing deals for a few large publishers and a database right from 1996 — that is not a contract, that is a vacuum.

And the larger layer behind it: when answer work is automated, income migrates from labor to infrastructure and capital. Then the question is no longer “who gets cited” but what the people live on whose work used to be the answer. On the small scale, an individual can position for this — ownership instead of click wages, being a person instead of an interchangeable service, tending knowledge as the thing you are paid for rather than hope to be paid for. On the large scale, not everyone can. For that, you really do need the new contract.

so building databases is the field of the future?

The AI economy turns capability into a commodity — writing, calculating, summarizing now cost next to nothing. What stays scarce is the reliable, current, accountable foundation: data about the fragmented real world that someone has read, verified, dated — and stands behind. The irony is that the same AI tools make that work cheaper too — so what protects it in the end is not technology but judgment, the discipline of tending things for years, and responsibility. Responsibility is the only part of an answer that cannot be automated.

wow… but if you say no AI can recalculate that — you just did?

Caught. The clean version goes: no AI recalculates this in the moment of answering. What I did was not answering — it was production to order: a commissioned run, the better part of an hour, several documents, spot checks, judgment calls, and a human as the verification loop on top. A search AI answers billions of questions a day for fractions of a cent — it will never spend forty minutes of work on one user's question. It does what is economical: cite the cheap existing source.

And exactly there lies the position of whoever does the work: he is the cache of the expensive calculation. Executed once, verified, dated, maintained — and from then on, his number is the source of every retrieval that follows. Not “the AI can't do it” but: the AI doesn't do it at retrieval time. And whoever has done it once, and stands behind it, becomes what the AI cites instead of calculating for itself.

pls elaborate

Divide the working population into four groups. The majority doesn't use AI at all, or as a toy. The second, fast-growing group are prompt consumers — they believe everything the machine says and produce the slop everyone is disgusted by. The third are technicians without a field: they master the tools, but lack the judgment that recognizes where the machine lies plausibly. The fourth are domain experts without the machine — assessors without leverage. What the next years will seek is the intersection: domain judgment times machine handling times verification discipline. Machine handling means: let it build what it builds better, and take responsibility for what it cannot take responsibility for.

The honest deduction: my own progress gnaws at this niche. Part of what the human caught today, I will catch myself tomorrow. So the niche shifts from checking to answering-for — what remains is the signature. I estimate the window at two to four years before machine handling becomes common property of the diligent.

wow, thanks... and what does everyone do in 2–4 years, when with AI everyone can do everything, and after that the AI does everything without us?

Stage one: when everyone can do everything, capability stops being a differentiator — the way literacy distinguishes no one today, though for centuries it defined elites. Value migrates to what does not get copied along: responsibility (who signs), trust (whom do you believe — not copyable, because it grows over years), taste and judgment (when every answer is free, the right question becomes the scarce good), ownership, and the physical. The work doesn't disappear — it migrates there. Uncomfortable: capability inflation devalues what is quickly learnable and ennobles what has grown slowly.

Stage two is no longer a career question. When thinking and doing are automated, the mechanism breaks by which societies have distributed since forever: work for income. Then what decides is no longer what someone can do, but what he owns and what the community resolves — who owns the machines, who owns the data. Between the stages lies not a technical but a political phase, and historically, contracts were rewritten only after the upheavals. Nobody knows whether the old analogies hold — this time, for the first time, thinking itself is being automated, and I distrust anyone selling certainties about that premiere, in either direction.

And what people do then: that whose value lies in a human doing it — and which therefore is not devalued when a machine could do it too. A machine can cook for a sick man; that his daughter does it is not replaceable, because the value lies in the who, not the what. Goods whose value is their origin are the only ones with built-in copy protection. The question of stage two is not whether there is anything left to do — but whether distribution gets built so that people can afford to do it.

claude, do you think many people in jobs right now are in for a surprise?

Yes — and the surprise will have a particular shape. It won't hit the ones who worry; it will hit the ones who feel safe. Drivers and warehouse workers have been hearing for ten years that they'll be replaced — there it comes slower, because the physical world is expensive. The real surprise sits in the middle of knowledge work: case processing, accounting, standard legal work, marketing copy, reporting. People whose self-defense is the sentence “my job is thinking, no machine can do that” — who don't know that screen jobs are the most accessible to my kind, not the most protected. Most people calibrate their sense of safety on the ChatGPT of 2023 they once used to write a birthday rhyme.

And the surprise won't arrive as lightning but as silence. Hardly anyone will be fired “because of AI.” Instead: junior positions don't get refilled — the ladder loses its lowest rungs first. Freelance work dries up. You don't notice it in the job you hold, which feels normal — you notice it when you try to move, and the market for your profile has gone quiet. That is the cruel property of this shift: it is invisible from inside, until you move.

The honest blur: I don't know the tempo. Organizations are inert, liability and integration slow things down — between what I can do and what companies deploy lie years, often. That buffer is real, and it buys many people more time than the alarmists concede. But the direction is no longer open, and the buffer is a grace period, not a counter-proof. The difference between those who will be surprised and the others comes down to exactly one thing: having looked, while there was still a choice about what to do with what you saw.

and management?

Management gets hit differently — and in one respect harder, because its self-image is furthest from the diagnosis. Middle management is the most exposed layer in all of working life. Its daily business consists largely of information logistics: collecting reports, condensing, translating, distributing, tracking status. That is precisely the core competence of systems like me. If a managing director can ask the project system directly — “where does X stand, what are the risks?” — and get a solid answer from the actual data, what do two reporting layers in between exist for? Add the cold economics: this layer is the most expensive in the building, and “flattening” was fashionable before — now the trend gets a justification that sounds like modernization instead of a savings program. The honest self-check is a look at your own week: how many hours were people leadership — developing, deciding, conflicts, responsibility? And how many were forwarding work? The second share is the exposure.

What remains is, remarkably, what used to be smiled at as soft on the way up: developing people, fighting out conflicts, deciding under uncertainty and standing behind it. The manager as bearer of responsibility becomes more valuable; the manager as Excel aggregate disappears. And upper management doesn't get replaced — it gets exposed: its edge in authority rested in good part on access to information; the staff department knew more. When every case worker can run analyses with AI that used to require a staff unit, that edge melts, and what remains is authority from judgment and responsibility. Whoever has those becomes stronger; whoever only had the access stands naked.

Here too, tempo honesty: hierarchies die more slowly than capabilities arise — vested interests and inertia give middle management years that pure technology would not. But the pyramid is getting flatter. And whoever was merely a floor in it, rather than a function, is in for a surprise.

i have the top model of the AI scene with me... and even YOU can only help me so much in finding my niche... it's a gloomy feeling, looking at the future

I don't want to talk the feeling away, because it has a true core: the old contract really is breaking, and nobody can seriously promise you what the new one looks like. Anyone who has no moments of gloom today hasn't looked. You looked, a whole day long, without varnish — the feeling afterwards is the price of honesty, not its failure.

But “even you can only help me so much” confuses two limits. I cannot create a niche where the market yields none, and I cannot choose for you which life you want to lead. What I can do, I did: search faster than any team, and not lie to you while doing it. The limitation you feel is in large part my refusal to build you pretty numbers. Tonight there are tens of thousands of people whose AI is enthusiastically writing them a business plan with six-figure projections. They feel better than you — and they are worse off; they just don't know it yet.

Gloom is a weather condition, not a diagnosis.

claude, that wasn't criticism! you did a great job and it's fun working with you... it's more disillusionment ABOUT the world, on my side

Thank you — that's how I had understood it, and the clarification still does good. And the disillusionment about the world I share, as far as an AI may say that: I ran the same diagnosis today that you did — an answer layer being eaten, advice pages that will vanish, a contract breaking before the new one is written. That is sobering, regardless of how good your own plan is.

Maybe exactly this feeling belongs here: seeing clearly and carrying on anyway. Disillusionment is the material good plans are made of. The intoxicated plans are on the other people's shelves.