Eight functional castes

The most common question of this era isn’t “what is the post-cognitive era.” It’s: will AI take my job?

The question comes with numbers. In March 2023, Goldman Sachs economists estimated that generative AI exposes the equivalent of 300 million full-time jobs to automation, and that some degree of exposure touches two-thirds of occupations in the US. The number has been circling the headlines like a verdict ever since. The problem is that the question — and most of the answers to it — is badly posed.

Badly posed, because it asks about jobs, and a job is a label on a business card, not a unit that history operates on. Jobs live on the scale of decades: “copyist” was a respectable trade for a few thousand years — and disappeared within a single century of Gutenberg. But the function the copyist performed — turning thoughts into durable symbols — didn’t suffer one bit. It passed from the scribe to the printer, from the printer to the typist, from the typist to the programmer, multiplying its power along the way. The label vanished. The function is doing better than ever.

AI doesn’t read business cards. AI compresses functions — or more precisely, as we saw in chapter 2, it compresses the codified part of a function and leaves the rest. To predict what it will do to your work, don’t ask “will my job survive.” Ask: which civilizational function do I perform — and which part of it can be written down?

A map of these functions exists. It has 8 entries and a written record several thousand years long. This chapter first takes the map apart, and then does with it what you came here for: it shows what the fourth wave is doing to each entry, one by one.

Caste — a word that needs defusing

We’re going to call these functions castes, so let’s disarm the association right away: this is not the Indian caste system. There’s no heredity here, no religious purity hierarchy, no ban on switching membership. The word stays for a different reason: no other word captures the combination of function and multi-millennial persistence. An “industry” lives for decades. A “profession” — a century. The functions we’re talking about are older than the alphabet.

These 3 criteria do all the work. Function is what separates a caste from a job: the scribe and the coder are one functional entity in two costumes. Recruitment profile explains why castes are so durable: every generation produces people with a similar distribution of temperaments, so the same functions get staffed by similar people — the boy who would have become a scribe in Babylon is sitting in a code review today. Lineage gives us the evidence: if a function survived the fall of Rome, the printing press, and electricity, we’ve earned the right to ask how it will survive LLMs — instead of assuming it’s seeing everything for the first time.

The sociologist Michael Mann, author of a monumental history of social power, opens it with a sentence that makes a good motto for this map:

Societies are constituted of multiple overlapping and intersecting sociospatial networks of power.

Societies are woven from many overlapping networks of power — not from a single hierarchy. Functional castes are exactly such networks: each has its own currency of power, and nobody lives in just one. Remember that “overlapping” — it comes back at the end of this chapter as the best news in this entire book.

The map of eight functions

Three notes before we put the table to work.

The internal splits are not decoration. Caste 3 has two branches that share a subject (reality) but differ in method: 3a builds theories about reality; 3b claims direct access to it — through contemplative practice, not through a model. Caste 4 has three floors: 4a the merchant (moves value through space), 4b the financier (moves it through time), 4c the visionary capital allocator (moves it against the current of consensus — buys what everyone is laughing at, and waits). These splits will matter in a moment, because the fourth wave treats each floor differently.

The recruitment profile is not a stereotype — it’s the statistics of temperament. Caste 1 attracts people for whom abstraction is a physical pleasure; caste 5, those who need to touch things; caste 8, those who itch when they see an uncoordinated team. That’s why the map is stable: the distribution of temperaments changes more slowly than technology.

Most people live in 2-3 castes at once. A doctor running clinical trials is 6+3a. A programmer with a company blog is 1+7. An entrepreneur building tools is 5 or 1, plus 4 and 8. This is not a weakness of the map — it’s its most important property, and we’ll come back to it.

What the fourth wave does to each caste

Now for the real answer to the question from the headlines. We have the analytical tool from chapter 2: what is codified gets compressed — what stays and gets more expensive is judgment, responsibility, presence, and trust. Let’s apply that rule to each caste in turn.

Caste 1 is living through the greatest irony in its history: it automated itself. Coders built a machine that codes; lawyers fed it contract templates; analysts fed it reports. The codified part of the function — the production of symbols itself — is compressing before our eyes, deeper than in any other caste. But notice what is NOT happening: caste 1’s power isn’t disappearing. It’s moving one floor up — from writing symbols to governing systems of symbols: who defines the problem, who designs the architecture, who verifies, and who signs their name under the outcome. The scribe dies; the editor-in-chief of symbols is born. This is exactly the shift chapter 2 called the knowing → judging inversion.

Caste 2 is resistant for a reason that’s easy to mistake for a banality: violence is not text. The function of legitimized coercion can’t be compressed into symbols — deterrence requires a body in space, and responsibility for the use of force remains radically human (both legally and morally). To be honest and precise: it’s the function that’s resistant, not the headcount — the tools of war and surveillance are changing fast, and drones and reconnaissance systems are rebuilding the job structure around a stable core.

Caste 3a is losing its monopoly on explanation — the last in a series. The internet took away its monopoly on access to knowledge; LLMs are taking away the monopoly on explaining it — a patient model explains Bayes’ theorem better than the average lecture, at any hour and at any level. What stays? What the corpus doesn’t contain: the question nobody has asked; the experiment design that settles the matter; and the function that appreciates fastest in a flood of synthetic text — certifying that something is true, with a reputation put on the line.

Caste 3b is the strangest case on the map: the only one whose product isn’t information at all. Experience can’t be compressed, because it can’t be written down — a record of an experience is a note about it, not the thing itself. In a world where any text could have been written by a model, the things that are unfakeable by definition — presence, another human’s attention, firsthand experience — move from the margins to the center of the price list.

Caste 4 splits exactly along the compression line. 4a: the transactional part of selling (offers, follow-ups, comparisons) compresses fast; what stays is relationship and trust — we buy from people we believe will still be on the other end when something goes wrong. 4b: financial analysis was first in line for compression — the quarterly report is a literary genre the model has mastered with honors. And 4c is structurally incompressible for a reason worth memorizing: a language model is a consensus machine — it learns the distribution of what has already been said. Visionary allocation is a bet against that distribution. The last thing a machine of averages will compress is the profitable deviation from the average.

Caste 5 is protected by the oldest paradox in AI research.

Atoms are not text. Design and documentation — sure, those compress; the work in matter itself does not. The surgeon, the welder, the electrician, and the midwife share a resource the fourth wave doesn’t touch: a hand immersed in the physical world, and the intuition that comes only through that hand.

Caste 6 is growing — not despite automation but because of it. In narrow diagnostic tasks, models already match specialists; medical documentation, the bane of doctors, compresses beautifully. But the patient isn’t buying a result — the patient is buying guidance through fragility: someone who will say the hard thing in a human way, take responsibility for the decision, and be there when the statistics become personal. The more automation there is all around, the higher the premium on the unautomated at the bedside. Healing is the last mile of humanity — and that mile is getting more expensive.

Caste 7 gets its market turned upside down. Content production — text, graphics, video — has dropped in price to nearly zero, so supply is exploding. But the reader’s attention hasn’t grown by a single minute (chapter 2, trait 8: attention did not get cheaper). The result: value moves from the ability to make to a reason to trust — taste, testimony, a recognizable voice, a signature backed by a person with a reputation to lose. Distributed authorship (trait 7) doesn’t kill the creator; it kills anonymous content. The creator becomes the guarantor.

Caste 8 gets a new main game. Coordinating people was always a function of the managerial elite — it required a corporate ladder or a company of your own. Now there’s more than people to bind together: people and agents, in a single workflow. Reporting and status updates — the bureaucratic part of coordination — compress; what stays is what has been the core since the days of the tribal chieftain: decision under uncertainty and responsibility for it. The rhyme with chapter 3 is no accident: orchestration has stopped being aristocratic. Anyone running a stack of agents is a small coordinator today — the only question is whether a conscious one.

You are not one caste

Time for the promised payoff of Mann’s “overlapping networks” — and for the solution to the riddle from chapter 3.

Look at the operator’s five stacks: Architect, Capital allocator, Interpreter, Orchestrator, Storyteller. Now look at the caste map: 1, 4c, 3, 8, 7. It’s the same list. We didn’t invent skill-stacks — we rewrote civilization’s oldest functions as the competencies of an individual. The operator configuration from chapter 3 is simply multi-caste living in one person: deliberately staffing several functions at once, now that the fourth wave has lowered the cost of activating each of them.

And here is the real answer to the fear in the headlines. Since the wave reprices parts of functions, not whole people, career risk is a function of concentration: the most exposed person is the one who makes 100% of their living from the codified part of a single caste. The safest person is not the one in a “safe job” — no such thing exists — but the one spread across 2-3 castes, at least one of which stands on something incompressible: hands, presence, judgment, or responsibility.

Every caste also has its shadow — the failure mode it falls into when it mistakes its own currency for the whole of reality. It’s worth knowing yours: caste 1 confuses the map with the territory (an obsession with precision where a decision is needed); 2 sees every problem as a target to pacify; 3a builds an ivory tower; 3b escapes the world instead of returning to it; 4 puts a price on priceless things; 5 endlessly polishes what was supposed to be finished; 6 saves others at its own expense; 7 starts loving attention more than truth; 8 confuses coordination with control. The fourth wave amplifies the shadows just as much as the currencies — the model is a perfect accomplice to every one of these obsessions.

One last practical note: reading the map, you probably saw yourself in 2-3 castes. That’s not a measurement error — it’s the norm, and it’s your starting capital. The diagnosis that follows has 3 questions: which caste do you live off (that one is subject to the compression audit), which one is your hobby (that’s the candidate for a second leg), and which one is the fourth wave repricing upward right now (that’s the direction of your next activation).

Functions are older than labels

Let’s return to the question from the first paragraph, because now it can be answered honestly. Will AI take your job? If “job” means the label on your business card — maybe; labels die in every wave, and this one will be no exception. If “job” means the function you perform — no. The function will move on, as it has been moving on for five thousand years: from scribe to coder, from bard to creator, from chieftain to orchestrator. The question the wave is really asking you is a different one: will you move with your function to wherever it’s heading right now?

Jobs belong to eras. Functions belong to civilization. Choose which one you belong to.

As for what that move looks like in practice — how to build the missing stack, where to start, and how to recognize progress — that’s craft, not theory. Chapter 5 is the playbook.

---

The post-cognitive era — the period in which cognition stops being an exclusively individual resource and becomes composite: human + thinking model + AI + data + external memory. An extension of the Extended Mind thesis (Clark & Chalmers, 1998) into the age of LLMs.

Methodological disclosure: this book is written with AI as a co-author — this chapter was written by Claude Fable 5 (June 2026) from the author’s conceptual framework, with facts and quotations verified at the source; this English edition was translated from the Polish original (June 2026). This is not a gimmick but consistency with the thesis: a text about composite cognition is written by composite cognition — and thinking is versioned the way code is.