Five skill-stacks

January, every gym in Poland: packed. February: half the crowd. March: the regulars and an echo. The difference between the people who stayed and the people who vanished almost never comes down to equipment or motivation — it comes down to one group having a training plan with reps and the other having a membership.

Skills work exactly the same way, just less visibly. Courses, books, subscriptions, and “read later” bookmarks are a membership. This chapter is the training plan: for each of the 5 skill-stacks — what you train, what the reps look like, how you’ll know it’s working. No more theory here: what an operator is, you know from chapters 1 and 3; where these five came from — chapter 4. You walk in with your test score from chapter 3 and walk out with a plan for your gaps.

One principle holds every routine below together, so let’s say it out loud. Chapter 2 showed that whatever is codified gets compressed — and what’s left is judgment, responsibility, presence, and taste. The training conclusion: don’t build a stack by warehousing knowledge. The model has the knowledge. Build through reps that cannot be written down: decisions with consequences, a public trail, contact with a reality that says “no.” Every routine in this chapter passes that test. If any of your current “learning” doesn’t — you’ve just found what to cut.

And one honest note about scope: I’m not promising 5 stacks in 5 weeks. I’m promising you’ll know what to train, and you’ll recognize progress when it arrives. Minimum horizon for the first signal: 90 days of routine. Anything faster is a placebo.

Architect — build to understand

What you train: thinking in layers (what sits above, what sits below, what talks to what), decomposing problems into parts that can actually be built, and reading other people’s systems — because the architect isn’t the one who can assemble things, but the one who knows why they fall apart.

This stack’s motto was on Feynman’s blackboard at Caltech the day he died:

What I cannot create, I do not understand.

In an era where the model will build almost anything for you, that sentence doesn’t expire — it changes address: what I cannot specify and verify, I do not understand.

Routines:

1. 1 working system per month. Small: an automation, a script, a household tool. The bar isn’t “it works when I demo it” — it’s “it runs for a week without me.” Build with the model, but write the spec yourself, before the first prompt goes out: the spec is exactly the part that doesn’t compress.
2. A layer walk, once a week. Take something you use — a mobile payment, a page of search results, the line at a doctor’s office — and map its layers on paper: what the user sees, what sits underneath, where the seams are, what happens when each layer fails. 15 minutes. This trains the eye, not the knowledge.
3. Read the autopsies. Public outage post-mortems (the world’s best engineering companies write them) are the cheapest architecture school in existence: systems teach the most where they cracked.

How you’ll know it’s working: when something breaks, your first hypothesis points at the right layer more often than every other time — and people start bringing you “where is this falling apart?” problems before they ask “how do we fix it?”

Capital allocator — train judgment, not forecasts

What you train: thinking in decades, spotting asymmetries (capped downside, uncapped upside), holding a position under social pressure — and the hardest part: knowing what not to do. Chapter 4 showed why this stack doesn’t compress: the model is a consensus machine, and allocation is a bet against the consensus.

Routines:

1. The decision journal. Every meaningful allocation — of money, but also of time and reputation — written down before the outcome: thesis, stake, review date, and one sentence on “what would convince me I’m wrong.” Quarterly review: you compare the record against reality, not a memory against reality. It’s the only known exercise that calibrates judgment instead of rationalizing it — memory will always rewrite the story in your favor; paper won’t.
2. The anti-task list. 3 popular things you deliberately do not do, each with a written justification. Refreshed every quarter. (You recognize this question from the test in chapter 3 — this is its training version.)
3. Train holding on small positions. Pick something with a horizon of at least a year — an investment, a project, a course of study — sized so the loss wouldn’t hurt, and don’t touch it. You’re not training the pick; you’re training not making moves: the rarest skill in an era where everything begs to be clicked.

Foundation reading (an exception, because this stack has a good library): “The Almanack of Naval Ravikant” (Jorgenson, 2020) on leverage and patience, “The Bitcoin Standard” (Ammous, 2018) on the hardness of money over time, “Antifragile” (Taleb, 2012) on systems that gain from shocks. Read once, they do nothing — read alongside your own decision journal, they make the difference.

How you’ll know it’s working: a growing share of decisions you judge the same way a year later as you did the day you made them; and a shrinking number of things bought, started, or promised in the heat of a single week.

Interpreter — collect patterns, not facts

What you train: reading outside your own field, carrying patterns across domains, and the rigor that separates an analogy from a proof. The model holds the facts; your edge is the bridges between fields that don’t exist in any corpus yet.

Routines:

1. The one-third diet. Every third book or long read — from outside your field, ideally from one with a completely different epistemology: history, biology, thermodynamics, Roman law. Not for erudition — for a stockpile of other people’s patterns.
2. The pattern note. After every serious read, one sentence in the file: the pattern + 2 domains it transfers to. “Natural selection = market test = product iteration.” A pattern file, not a quote file — after a year you have a private dictionary of bridges nobody else owns.
3. The weekly steelman. Once a week, build the strongest possible version of a view you disagree with. The model is the perfect sparring partner here — make it defend the position for real and don’t let go until you feel your certainty wobble. If it hasn’t wobbled once in a month, you’re not training — you’re cheering for yourself.

How you’ll know it’s working: “this is the same thing as…” moments show up more and more often in new situations; and people start saying “I’d never have thought to connect those.”

Orchestrator — close, don’t open

What you train: decomposing work into parallel streams, a review cadence, delegation with quality control — these days not just to people but to agents — and the most underrated skill of the era: closing things out.

Routines:

1. 2-3 streams with a weekly review. A professional project, a personal project, one extra. Once a week, 30 minutes, in writing: what moved, what’s blocked, what’s next. Not in your head — in writing; orchestration in your head is juggling, orchestration on paper is a system.
2. 1 delegation a day to an agent. Every day, hand the model or an agent one task with a clear acceptance criterion — and log what came back good and what needed fixing. It’s the cheapest delegation training in history: you calibrate your specs without risking anyone else’s time. Whoever learns to accept work from agents today will know how to accept it from hybrid teams tomorrow — chapter 4 called this caste 8’s new main game.
3. The WIP limit. Set a maximum number of parallel commitments. A new one gets in only when an old one is closed or explicitly killed. A thing abandoned without a decision doesn’t disappear — it hangs there, collecting rent from your attention.

How you’ll know it’s working: the ratio of things finished to things started goes up; and nobody has to ask you “so, what’s happening with that?” — the status comes from you before the question exists.

Storyteller — publish on a cadence, not on inspiration

What you train: clarity (writing is thinking held up to the light), a publishing cadence, a bond with a specific reader, and taste. Chapter 4 showed the stakes: content production got cheap to the point of zero, so the entire premium moved to the reason to trust — voice, testimony, signature.

Routines:

1. 1 public piece a week, for a year. Not for reach — for 52 reps with feedback from reality. Short is fine; unpublished is not. Publishing isn’t vanity, it’s a training condition: a piece written for the drawer costs nothing, and only the reps that cost something train you.
2. The smart 12-year-old test. Once a week, take one concept from your work and explain it so a smart 12-year-old would get it — without lying by simplification. This isn’t an exercise in simplifying; it’s a merciless test of whether you understand it yourself. (This entire book is written under that rule — the shopping mall in chapter 1 is the test in action.)
3. Read aloud and study the masters. Read every piece out loud before publishing — the ear catches what the eye can’t see. Once a quarter, take apart one piece by a writer who steers attention masterfully: not what they wrote, but in what order and what they left out.

And the model? Editor, critic, counterargument generator — yes. Ghostwriter — no, and not for reasons of honor: a piece written entirely by the model trains the model, not you, and the only thing getting more expensive in caste 7 is precisely what the model can’t sign.

How you’ll know it’s working: people start writing back and passing your pieces along; and the “can you explain this in plain words?” requests start arriving on their own, uninvited.

Where to start — the build order

Your test score from chapter 3 tells you what’s missing. The order in which you fill the gaps has a logic of its own:

Start the decision journal today, no matter what. Capital allocator calibrates over years — the journal’s value grows with the age of its entries, so every month of delay is a month of calibration lost for good. It’s 20 minutes a week; there is no cheap excuse for it not to exist.

Interpreter is the cheapest entry point. It starts with changing your reading diet — technically: tonight.

Storyteller compounds earliest. A public trail builds the other stacks on the side: you write about what you’re building (Architect, documented), about what you’re not doing (Allocator, said out loud), about the patterns you connect (Interpreter, in writing). One piece a week is scaffolding for the whole configuration.

Architect needs projects, Orchestrator needs fronts. These two grow from use, not from reading; they enter naturally once the previous three start feeding you things to build and tie together.

And the master rule: 1 new stack at a time, 90 days of routine before judging. Trying to build three at once is the January gym — we know how that ends.

One last thing — the synergy chessboard, because stacks multiply rather than add: Architect × Interpreter makes a systems thinker (Vitalik’s configuration from chapter 3); Capital allocator × Orchestrator makes a founder; Storyteller × Interpreter makes an essayist — the configuration of Naval, or of Tim Urban, whose style this book openly borrows. When picking your second stack, don’t ask “which is best” — ask “which multiplies my first.”

The price of a personal protocol

Chapter 3 called the operator a personal protocol: value that lives in the rules for connecting layers and survives the replacement of every single one. This chapter showed the other side of that definition. A configuration cannot be bought, read, or generated — it can only be trained, through reps that cannot be delegated. The good news: every routine in this chapter starts this week. The uncomfortable news: none of them ever ends. That is the price of a personal protocol — and the exact reason it cannot be copied.

You now have the map of the era (chapters 1-2), the diagnosis (3), the origin story (4), and the training plan (5). One piece of the personal puzzle remains: relationships. Your configuration doesn’t work in a vacuum — it works in relationships with models, agents, machines, and systems, and those relationships have their own architecture and their own failure modes. Chapter 6 maps 4 types of these relationships plus a fifth, the most interesting one — triads — and will help you find the weakest edge of your own.

---

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 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.