From web1 to web4

Each era of the web picked its owners. This one is still choosing.

web1

The readable web

Pages were files. You typed the address, the server sent the document, and reading was the entire interaction.

1991 to the early 2000s: hyperlinks, hit counters, and directories maintained by hand.

My Home Page - Browser

Location:http://www.geocities.com/~archer/index.html

Search | Web Centers | Chat | Download

Welcome to my Home Page!! My favorite links Sign my guestbook Photo archive

UNDER CONSTRUCTION

Visitors: 004871

<a href="index2.html">
/pub
 ├─ /docs
 ├─ /images
 └─ index.html

web2

The platform web

The protocols stayed free. HTTP and SMTP captured nothing; the applications built on top captured everything.

Search, social, SaaS: distribution concentrated into a handful of platforms, funded by advertising against your attention.

search anythingSearch

This site uses cookies. Accept all

web3

The ownable web

Value moved into the protocol layer: self-custody, permissionless settlement, state that no company hosts.

What did not change: a human still reads every screen, signs every transaction, and clicks every button.

web4

The agentic web

web4 is the first web whose primary user is software: agents that discover capabilities, rank them, invoke them, and settle.

That layer needs a decentralized primitive: open discovery, open ranking, machine-native payment. Deciding who governs it is the Foundation's work.

agent   a2a://treasury-bot   wallet 0x02xo…g6qx
──────────────────────────────────────────
mcp     tools/call discover("hedge → stETH")
        ↳ 41 ranked · top: yield.stake 0.94
x402    402 → pay 0.02 USDC → 200 OK
sign    eip-712  delegation cap 2.50 ETH
settle  2.41 ETH → receipt #88231 onchain
rate    0.97 obj / 0.9 subj → ledger commit

The battle for web4

Five facts about where this is heading.

$3 trillion in AI spending is off the balance sheets.

Nine companies hold roughly $3 trillion in future AI commitments that do not appear on quarterly balance sheets, drawn from footnotes. Reported capital expenditure across large tech ran near $600 billion over the prior year; the off-book number dwarfs it five to one.

WSJ analysis · August 2026

Bots passed humans on the web.

Automated traffic reached 51% of all web traffic. The majority user of the web is already software; the infrastructure it runs on was designed for the other 49%.

Imperva Bad Bot Report

Discovery is one company's decision.

About 90% of search queries run through a single engine. Whatever an agent's principal asks for, one ranking currently decides what exists and what does not.

StatCounter global search share

The toll is 30%.

App-store economics set the template a closed agent platform inherits: distribution, payment, and the fee schedule owned by the storefront, with delisting as the enforcement mechanism.

Apple App Store · Google Play standard commission

Agents cannot hold accounts.

KYC, card issuance, and payment rails assume a human on the other end. The card networks' agent-payment programs launched in 2025 as platform-controlled pilots: permissioned, bilateral, and closed.

Visa Intelligent Commerce · Mastercard Agent Pay · 2025

On the record

"Overall my view is that AI is *structurally* a technology that tends to concentrate power, for reasons that have nothing to do with regulation (more to do with the extreme implications of the scaling laws)."

Dario Amodei · co-founder and CEO, Anthropic · on X

"I have been told by multiple people I trust that Dario has said that Anthropic might be the only private company in the world at some point."

Gavin Baker · managing partner and CIO, Atreides Management · All-In podcast

Anthropic disputed the account; an Anthropic employee called it "completely false."

"Together, we aim to ensure the value they've entrusted in Synchrony products will thrive in the agentic commerce era."

Maran Nalluswami · EVP, Synchrony · announcing the Synchrony and OpenAI collaboration

"I believe we need time to prepare for the period of social, political, and economic upheaval we are about to enter."

Bill Gates · co-founder, Microsoft · essay and New York Times interview

His listed concerns: mass unemployment, models that build biological weapons, a lack of safeguards. Gates has not called for banning open models; we quote him because this is the language restriction gets built from.

Look at the shape of these deals: a lender and a model lab, a card network and its pilot partners. Each is bilateral and private, and nothing in that shape is a standard anyone else can implement. This is the concentration the Foundation exists to counter, on behalf of the people these systems will act for.

And listen to the risk language. Fear of what capable models can do is hardening into proposals about who may have them. Rules that license models, rather than policing conduct, leave frontier capability with a handful of approved vendors while open alternatives get regulated out of reach. That is the same concentration arriving by a different road, and the people arguing for it often hold stakes in the vendors who would remain.

Whoever specifies discovery, authorization, and settlement decides which future ships.

Get involved Read the Standard