Protocols in a shopping mall

Wednesday, 3:07 p.m. The M1 shopping mall in Zabrze, Poland.

The doors open before you reach them — a motion sensor picked you up a few steps back. Nobody promised you this; there’s no sign that says “doors respond to approach.” You just walk in.

At the cart corral you drop a 2-zloty coin into the lock and the chain lets go. Stay with this for 10 seconds, because it’s a small miracle of design: nobody guards the carts, nobody writes down who took which one — and the carts come back. Someone invented the rules once (a deposit, a lock, a chain), and for decades millions of carts in thousands of stores have been returning themselves, without a single “cart watcher” on any payroll. A rule written once, executed billions of times.

At the checkout you line up on the left side of the belt. Why the left? Because the belt moves purchases from left to right relative to the cashier: in on the left, out on the right. Nobody ever explained this to you. It took 30 seconds of watching — and the protocol was in your head.

At the exit stands a security guard who doesn’t check receipts. His presence is not a control procedure — it’s a deterrence protocol. A procedure works every single time; deterrence works statistically. Both are rules of the game — of different kinds.

Within 10 minutes at the M1 mall you’re operating a dozen-odd protocols — technical, economic, legal, social — and not one of them has passed through your conscious mind. Your brain intercepted them, cataloged them, and is executing them. You don’t ask why the cart has a coin lock. You operate.

This is not a metaphor. This is how every complex human system works, from a shopping mall to a civilization. That’s why we’re starting in Zabrze — and we’ll end at the question this entire book stands on: what happens when protocols stop being about carts and money, and start being about thinking itself?

This book claims that this is exactly what’s happening — that we are living in the first years of the post-cognitive era , an epoch in which thinking stops being an exclusively individual resource. I know how the term sounds. It sounds like a conference slide. So before you close the tab: I didn’t coin “post-cognitive era.” Let’s start in 1998, when two philosophers asked: where does the mind end?

That year, Andy Clark and David Chalmers published a paper in the journal Analysis titled “The Extended Mind.” The thesis was provocative then and is getting harder to dismiss today: the boundary between “mind” and “environment” is a convention. For centuries, philosophy assumed that thinking happens in the head — that the skull is the border of cognition. Clark and Chalmers attacked that assumption with a single thought experiment.

Inga and Otto, two New Yorkers, want to see an exhibition at the Museum of Modern Art. Inga thinks for a moment, remembers — 53rd Street — and goes. Otto has Alzheimer’s and doesn’t retain current information in memory, but he carries a notebook everywhere and writes down what is where. Otto opens the notebook, reads: museum, 53rd Street — and goes.

Clark and Chalmers’s question: what is the functional difference? Inga pulled the address from biological memory, Otto from paper memory. Both pieces of information were waiting, ready, before the question was asked; both led their owner to the destination; both meet the criteria by which philosophy of mind defines a “belief.” Why should we say that Inga thinks while Otto merely uses a tool?

The answer: there is no good reason. Otto’s notebook is part of his cognitive system — not an external helper but an external component of his mind. The principle they derived from this is called the parity principle: if something outside the head does exactly what we would, without hesitation, call a cognitive process inside the head, then it is part of the cognitive process. Function counts, not location.

Cognitive processes ain’t all in the head. The environment has an active role in driving cognitive processes.

Clark and Chalmers were not claiming, mind you, that everything is mind. They set conditions: the external resource must be constantly available, immediately accessible, and automatically trusted — like the notebook Otto always has on him and trusts without checking. Remember those 3 conditions. In a moment we’ll hold them up against the device sitting in your pocket right now.

Because the line of thought didn’t stop in 1998. In 2008, Clark expanded it into the book Supersizing the Mind, and Chalmers wrote its foreword, stating flatly: “The iPhone is part of my mind already.” In 2011, Google Maps took over the spatial navigation of hundreds of millions of people: constantly available, immediately accessible, automatically trusted — all 3 conditions met more strongly than Otto’s notebook ever met them. In the meantime, the extended mind thesis grew into an entire school of cognitive science (so-called 4E cognition). And in December 2025, Clark himself closed the loop: in Nature Communications he published “Extending Minds with Generative AI,” describing generative AI not as an alien agent but as the next layer of a system that was distributed all along — brain + body + environment.

So the thesis of this book — cognition has become composite: a human + their thinking model + their AI + their data + their external memory — is not a hot take. It is the extension of a 28-year philosophical line that its own author carried into the age of LLMs.

Hence the name. “Post-cognitive era” doesn’t mean “the era after thinking” — just as “postindustrial” doesn’t mean a world without factories. It means the era after exclusively-individual cognition: the time when Clark and Chalmers’s thesis stopped being a thought experiment and became the literal infrastructure of everyday life — and when the very mechanism of extending the mind came under protocol standardization. What the second half of that sentence means, you’ll see by the end of this chapter.

The four waves of extension — why “post” and why now

A critic has every right to ask: we’ve been extending the mind for 7,000 years, so why are you announcing a new era only now? Fair question. The answer fits in one table.

The first 3 waves extended memory and access to information. Wave 4 extends the processing itself. Writing: “remember this for me.” Print: “hand this out to others.” Internet: “find this for me.” AI: “think this through with me.”

This is a difference of category, not degree. Before the microscope, we had ever-better instruments for looking — eyeglasses, spyglasses, binoculars. The microscope was not a better pair of binoculars: it opened a category of observation that hadn’t existed before, and out of it grew microbiology and histology. In the same way, an LLM is not a better search engine. A search engine serves up information; the model reasons together with you — and that is an activity which, for the entire history of the species, happened exclusively inside skulls.

The effect shows up most sharply in expertise. The 10,000-hour rule doesn’t disappear, but some of those hours change address: they move from the expert’s head into the tool. A junior with a well-configured AI stack doesn’t suddenly become a senior — but they do a growing share of the work that still required a senior in 2022. Expertise stops working as the entry barrier it used to be. (What exactly compresses, and what stubbornly doesn’t — that’s chapter 2; the distinction will turn out to matter more than it sounds.)

The fourth wave is why the “post” is justified. But to see how this wave spills outward — and what you’re supposed to do about it — you need the concept from this chapter’s title. Back to the shopping mall.

What a protocol is and why this word carries the book

The word “protocol” has been worn thin: there’s diplomatic protocol (etiquette), medical protocol (procedure), the TCP/IP protocol (a technical standard). We need a sharp definition, because the rest of the argument stands on it.

3 elements of this definition are doing the work:

Agreed — someone established these rules and someone accepted them: through an ISO standard, through social evolution (the queue), through cryptographic code (Bitcoin), through a technical specification (MCP). How the agreement happened determines the protocol’s character.

2 or more parties — a protocol always describes an interaction. You can’t “do a protocol” alone; it is relational by definition.

Without a coordinator at every interaction — this is where the heart of it sits. The rules may have been written by someone central (evacuation regulations are written by a legislature), but the execution happens without them: nobody stands by the carts conducting the returns, no “Internet Central” steers packets between routers. The coordination was written into the rules once — and from then on it executes itself, billions of times. A procedure is a manager on duty. A protocol is the manager who wrote down the rules and went home. Forever.

Protocol vs application — the most important distinction in this book

The ATM at the entrance to the M1 mall is an application: a specific terminal, a specific brand, a specific implementation. But the fact that your card from a Polish bank also works in an ATM in Tokyo is guaranteed by a protocol: ISO 8583, the standard for financial transaction messaging published in 1987. The ATM’s operator can go bankrupt, the terminal can break down — ISO 8583 will remain.

Applications have owners, and owners change their minds, go bankrupt, get acquired. Protocols, once adopted widely enough, become infrastructure — durable like roads, hard to replace like the alphabet.

The entire history of technology repeats this layering. TCP/IP (1974) is a protocol — websites are applications on top. SMTP (1982) is a protocol — Gmail is an application. Bitcoin (2009) is a protocol — crypto exchanges are applications. MCP (2024) is a protocol — specific AI tools are applications.

Applications fight over the market. Protocols become the market.

One caveat, before anyone reads this as “go build protocols” advice: a protocol rarely pays a dividend. TCP/IP made no one a billionaire — the billionaires were the people who understood earliest what TCP/IP would do to commerce, media, and advertising, and rebuilt their businesses around it. Bezos didn’t build a protocol; he built an application on a protocol before the rest of the world noticed that the rules of the game had changed. Value rarely settles in the protocol itself. It settles with those who read the new rules first.

3 types of protocols — you saw all of them at the M1 mall

Technical — the door sensor, the payment terminal, the cart lock. Enforced by physical infrastructure: they work whether or not you understand them.

Institutional — opening hours, evacuation signs, safety signage. An institution writes the rules, but nobody in the store conducts the evacuation — the sign works on its own. Here, ignorance costs you: “I didn’t know” doesn’t exempt you from the regulation.

Social — the checkout queue, “stand on the right, walk on the left,” eye contact with the security guard. Enforced by social pressure; they work statistically, not deterministically. Cut in line — nobody arrests you, but the system of stares kicks in.

All 3 types coexist, and together they make a shopping mall run without any control center coordinating each interaction. Exactly the same thing is happening one floor up, at the level of civilization — except instead of coin locks we have financial protocols, communication protocols and, recently, cognitive ones. Two of them you need to meet up close, because the rest of the book stands on them.

Bitcoin (2009) — a protocol instead of a bank

On October 31, 2008, someone signing as “Satoshi Nakamoto” sent a link to a PDF to a cryptography mailing list: “Bitcoin: A Peer-to-Peer Electronic Cash System.” 9 pages, 8 references. The thesis, in the first sentence of the abstract: a purely peer-to-peer version of electronic cash would allow payments to be sent directly between parties, without the involvement of a financial institution.

Sounds innocent. To see the revolution, you have to see why this was impossible for the entire earlier history of money.

Double-spend: why digital cash didn’t exist before 2009

A digital file can be copied. If money is a file, it can be spent 2 times — that’s the “double-spend problem,” for decades the fundamental barrier to digital cash.

Every solution before Bitcoin came down to a central ledger: the bank remembers that Jan has 100 zloty and has already spent 80. Visa, PayPal, clearing houses — different signs over the door, the same model: a trusted third party keeps the books and settles disputes.

And a trusted third party is power. It can freeze an account, refuse a transaction, charge a fee for being in the middle, carry out a politically ordered block — or go bankrupt with your money inside.

Bitcoin solved double-spend without a trusted third party. The mechanics in 4 sentences: every transaction is announced publicly to the entire network. Nodes compete computationally (“mining”) for the right to append the next block of transactions to the shared history. Each block is cryptographically linked to the previous one — hence “blockchain.” Faking a historical transaction would require recomputing its block and all the following ones faster than the rest of the network adds new ones — that is, permanently outrunning the combined computing power of all other participants at once: theoretically possible, economically absurd.

Remember the cart and the 2-zloty coin? It’s the same design move, raised to a higher power. The protocol doesn’t assume people are honest — it sets the incentives so that honesty pays and cheating doesn’t. The deposit in the cart lock and the reward for a mined block are one idea at 2 scales: economics built into the rules instead of an overseer standing over the rules.

The result: you can send value to a stranger on the other side of the world, with no intermediary at all, and both sides can treat the transaction as practically final — because it was validated by a protocol, not an institution.

Whether Bitcoin is a good currency is a separate, contentious dispute (volatility, energy, adoption). But Bitcoin as a settlement protocol is a structural claim, not an investment claim: for the first time in history, a financial transaction settles without a central arbiter.

The pattern, not the particulars

For this book it doesn’t matter whether Bitcoin “wins.” What matters is the pattern:

1. There was a problem that for centuries required a central intermediary (double-spend).
2. Someone designed a protocol that solves it with rules — cryptography and incentives.
3. The protocol is permissionless: you don’t need anyone’s permission to participate.
4. The protocol is trustless: you don’t trust people, you trust the math.
5. The result is practically irreversible once the rules have been satisfied.

Remember this pattern — a protocol instead of an intermediary — because in 2024 it showed up again. Not in finance. In cognition.

MCP (2024) — a protocol instead of a gatekeeper

On November 25, 2024, Anthropic published the specification of the Model Context Protocol (MCP): an open standard for communication between AI models and tools, data, and other agents. (Honesty compels a note: Anthropic is also the maker of Claude — the AI co-writing this book. Treat this as a declared conflict of interest, and judge the argument, not the author.)

As usual with protocol breakthroughs, the technical description sounds boring. Let’s look at the problem MCP solves.

N×M — why AI integrations were a walled garden

You have 3 AI models and 5 tools — GitHub, Slack, a database, a payment system, a CRM — and you want every model to be able to use every tool. Every connection is a separate integration: 3 × 5 = 15 implementations, each with a different API and different authentication. At enterprise scale — 20 models, 100 tools — that’s 2,000 integrations. Unmaintainable.

Before MCP, every big player solved this with a walled garden: OpenAI had its own tool-calling format just for its own models, Google had its own for Gemini. You want to be a tool in ecosystem X — you integrate with X, on X’s terms, with X’s permission. This is exactly what computer networks looked like before TCP/IP: IBM’s network didn’t talk to DEC’s network; every manufacturer had its own closed protocol.

MCP does to this what TCP/IP did to networks: a common communication layer. 1 implementation on the model’s side, 1 on the tool’s side — and everyone talks to everyone. N×M becomes N+M.

Capability discovery — the deeper difference

There is also something the transport protocols never had: capability discovery. An agent connects to an MCP server and asks: what can you do? It gets a structured answer: these tools, this access to resources, these commands. It can discover what the environment makes possible before it sends its first request.

This is the codified reflex of an expert at a new workplace — not “how do I transmit data” but “what can be done here?” You know this firsthand, by the way: your brain spent its first 2 minutes at the M1 mall doing exactly the same thing — scanning which checkouts were open, where the carts were, whether the food court was running. Capability discovery. Only then came the plan.

Proof by adoption — the TCP/IP moment in 12 months

The strongest proof that MCP is a protocol and not a product came from the outside. In March 2025, OpenAI — Anthropic’s main rival — announced MCP support in its products. In April 2025, Google announced support in Gemini, and the head of Google DeepMind called MCP “rapidly becoming an open standard for the agentic AI era.” In May 2025, Microsoft built MCP into Windows 11 as a system-level layer for agents. And in December 2025, Anthropic handed MCP over to the Linux Foundation — the neutral organization that stewards, among other things, the Linux kernel — relinquishing control of its own standard.

Competitors don’t adopt each other’s products. They adopt protocols. The walled gardens agreed on a shared gate — this is exactly the TCP/IP moment, except there it took 2 decades and here it took 12 months.

What MCP shares with Bitcoin — and what it doesn’t

Let’s be precise, because this pairing is easy to abuse. Bitcoin is trustless: math replaces trust in the counterparty. MCP is not trustless — you have to trust the server you plug in, and agent security is still an open engineering problem. What they share is a different, more important element of the pattern: permissionless, with no central owner. Nobody issues permission to implement MCP and nobody can shut you out of it — just as nobody issues permission for a Bitcoin transaction.

In the shortest form: Bitcoin removed the mandatory middleman from finance. MCP removed the mandatory gatekeeper from machine cognition. 2 domains, 1 move — a protocol instead of a point of control.

And here everything clicks together. In 1998, Clark and Chalmers described how a single mind extends itself with a notebook. MCP standardizes the edges along which that extension flows today: human + agent + tools + data + memory compose into one composite cognitive system — and no single owner controls the rules of that composition. The extended mind stopped being a philosophical thesis. It got a specification.

This is the post-cognitive era at the level of protocol.

The protocol-era operator — you, the reader

We now have the pieces: the shopping mall as a network of invisible protocols; the fourth wave of extension as a qualitative threshold; Bitcoin and MCP as protocols that remove central points of control — from finance and from cognition. One piece is missing: someone who sees all this and knows what to do with it.

In a shopping mall there is usually 1 person who understands all the protocols at once: the mall manager. They know why the coffee shop sits at the entrance (it anchors dwell time), why the food court is far from the entrance (it pulls traffic through the whole mall), why the carts have coin locks. The customer sees stores. The manager sees a network of protocols for flow, time, and conversion — and optimizes it as a system. Same building, 2 levels of operating.

The same split repeats in every technological era. In 1991, most people saw “web pages”; Tim Berners-Lee gave HTTP and HTML to the world with no patent and no fees, because he understood he was building a protocol, not a product. In 2008, investors saw “new money”; Satoshi was designing a settlement protocol. In 2024, most people see “a chatbot”; the protocol-era operator sees that the infrastructure of cognition is just now getting shared standards — and understands what that changes structurally.

What a protocol-era operator is NOT

The term sounds technical, so it’s easy to get this wrong. 4 boundary lines:

Not an AI expert. An AI expert can fine-tune models and optimize prompts — application-level competencies, valuable ones. An operator understands how AI protocols rearrange the structure of power and the flows of value. That’s a different level of abstraction, not necessarily a technical one.

Not an early adopter. An early adopter has the new gadget on launch day. An operator doesn’t have to be first — they have to understand what is a protocol and what is merely a novelty. You could have had an email account in 1995 and never noticed that SMTP was just then taking the monopoly on business communication away from fax and the postal service.

Not a crypto enthusiast. A maximalist optimizes for 1 protocol. An operator treats Bitcoin as a pattern — permissionless, trustless, incentives instead of an overseer — and looks for that pattern in the waves that follow, indifferent to tribal colors.

Doesn’t have to be a programmer. You can understand TCP/IP without being able to write a packet sniffer, and MCP without knowing the spec by heart. Architectural understanding is a different thing from implementation skill.

Historical precedents

This is not a new type of human — it’s an old pattern in a new costume.

Florence, 15th century. The Medici didn’t invent the bill of exchange or double-entry bookkeeping — but they standardized those tools and stretched them across a network of branches from London to Naples. A merchant who understood how the Medici system worked could trade across borders; one who only knew how to trade locally fell behind when the scale of European commerce exploded.

The East India Company, 17th-18th century. The Company’s most effective people didn’t manage ships and pepper — they managed protocols: trade treaties, relationships with local rulers, the seasonality of the winds, and the rotation of capital. (An operational pattern, not a moral one — this was an empire of monopoly and violence.)

Silicon Valley, 1995-2005. Thousands of companies were “doing the internet” — that is, building prettier portals. The most durable companies of the era were built by those who understood that TCP/IP and HTTP were new infrastructure through which every industry would have to rebuild itself — and that whoever read those rules earlier had a decade of advantage.

The pattern is constant: whoever understands a protocol before it becomes obvious holds a structural advantage — the kind that can’t be quickly copied, because competitors would first have to see what they don’t see.

The 5 skill stacks — a sketch

Chapter 3 expands this into a full portrait with a self-test. Here, just a sketch so you know where we’re headed. A protocol-era operator combines 4-5 of the 5 stacks:

Architect — understands systems from the inside, at the design level; knows how a protocol works, even without implementing every layer.

Capital allocator — allocates time, attention, and capital on a horizon of decades; understands that an investment in understanding a protocol has a different return profile than a bet on a single application.

Interpreter — reads reality through several lenses at once: technical, historical, economic, philosophical; recognizes recurring patterns — “this is another Bitcoin moment, just in a different domain.”

Orchestrator — coordinates many projects and relationships without losing coherence; in the protocol era, value often arises at the seams between domains, and seams demand coordination, not narrow expertise.

Storyteller — translates complex patterns into public language. A protocol becomes infrastructure only through adoption, and adoption requires a story: Satoshi wrote a whitepaper, not just code.

No single stack makes an operator — a configuration of 4-5 does. Naval Ravikant: Interpreter + Capital allocator + Storyteller. Vitalik Buterin: Architect + Interpreter + Storyteller. Satoshi Nakamoto (as a phenomenon, whoever they were): Architect + Interpreter + Storyteller. I’m not invoking them because they’re famous — I’m invoking them because they are publicly verifiable examples of this configuration of competencies.

The map ahead

You now have 3 things, and they’re enough to read on.

First, a precise concept of a protocol: agreed rules that let parties interact without a coordinator at every interaction. Protocols endure, applications pass.

Second, the pattern in 2 acts: Bitcoin removed the mandatory middleman from finance (2009), MCP removed the mandatory gatekeeper from machine cognition (2024) — and within 12 months it was adopted by the competition, then handed to a neutral foundation. This move — a protocol instead of a point of control — will not stop at 2 domains.

Third, the term’s pedigree: “post-cognitive era” is not a buzzword but the extension of a 28-year philosophical line — from “The Extended Mind” (1998) through Chalmers’s iPhone (2008) to Clark’s own “Extending Minds with Generative AI” (2025). The era after exclusively-individual cognition.

From here, the book goes like this:

Chapter 2 defines the post-cognitive era operationally: the full table of the 4 waves, 8 traits of the era — from asymmetric cognition to the hyperliquidity of intent — and the Floridi-Stiegler-Clark philosophical line.

Chapter 3 takes the protocol-era operator apart: why configuration wins over specialization, the historical precedents, and the “is this you?” test.

Chapter 4 answers the question “will AI take my job” — but at the right level: 8 functional castes, their millennia-long lineages, and what the fourth wave does to each of them. Chapter 5 is pure practice: how to build each of the 5 skill stacks.

Chapter 6 sorts out the 4 types of relationship with AI — from human↔AI to human+AI+machine triads — along with the failure modes of each.

Chapter 7 makes the case for Bitcoin as Cryptographic Power — a 5th category of social power alongside Michael Mann’s 4.

Chapter 8 measures the 2023-2030 window: why epochal windows last 5-15 years, how to tell a window from a bubble before history answers for you — and what this window looks like from Poland.

Chapter 9 lays out the 5 risks of the era: cognitive atrophy, homogenization of thought, manipulation through control of the models, fragility of infrastructure, stratification of access.

Read in order, or navigate by the glossary; the full table of contents is waiting on the home page.

Most people will pass through the post-cognitive era the way they pass through the M1 mall: smoothly operating protocols they never see. This book is for those who want to see. Because to see a protocol before it becomes invisible — that is an advantage you can’t buy later.

---

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 — successive versions of its chapters are written and revised by successive generations of models (first version: Claude Opus 4.7, May 2026; current revision: Claude Fable 5, June 2026 — deeper argumentation, verified sources, corrected facts; this English edition 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.