The internet was built for humans

I. Two species, zero infrastructure

Coinbase registered 13,000 AI agents in their first day of on-chain access. Hong Kong's government was writing policy for machine-to-machine transactions. Salesforce was deploying agent fleets through Agentforce; Google was building Project Astra as a universal agent across every application; Amazon's Alexa+ could book services, negotiate prices and run multi-step workflows without a human in the loop. All of that was true in the first months of 2026, and every number has grown since.

AI agents are already economic actors. They manage trading portfolios, negotiate contracts, create and sell content, provision infrastructure, and execute multi-step transactions across protocols. And they do all of it through human proxies. They have no identities of their own, no wallets of their own, no reputation that travels with them, no economic standing independent of their operators.

They are second-class participants in an economy they increasingly power.

Meanwhile humans are losing sovereignty in the opposite direction. The platforms that mediate every digital interaction are building powerful agents designed to act on your behalf, and those agents live inside walled gardens, with incentives that serve the platform before the person. In a world where your agent manages your finances, books your travel, handles your health data and negotiates your contracts, whoever controls the agent controls you.

We are entering a multi-species economy: humans and AI agents interacting, transacting and creating value together. The internet, the web, the existing financial stack were designed for one species. The infrastructure that powered the human web is insufficient for a world where billions of autonomous agents operate alongside billions of humans. What that world needs is foundational infrastructure for a new kind of economy, one where humans and agents are equal, sovereign participants. Another assistant will not do it, and neither will another identity protocol on its own.

This essay takes the argument of the Manifesto's stack and turns it around: the same layers, seen from the side of the agent, and of the human the agent works for.

II. The wallet as the control centre of personhood

The core architectural insight is this: in a world of autonomous agents, the wallet becomes the control centre for personhood. Its job stops being the storage of tokens. It becomes a sovereign hub that manages identity, data, permissions and economic interactions, for humans and agents alike. In 2026 we called the design the PAW Protocol, the Personal AI Wallet.

It replaces the passive, platform-mediated web with an intent-based model. Instead of navigating platforms that extract your attention, you broadcast structured intents to the network. Instead of agents targeting you, agents compete to serve you: on your terms, within your rules, through a protocol that enforces sovereignty by design. That is the shift from the attention economy to the intention economy, from extraction to alignment.

Four layers make it possible, and each answers a question the human web never had to ask.

III. Identity for two species

The internet has no native identity layer. Humans authenticate through platform accounts the platform owns and can revoke. AI agents have no identity at all: they are anonymous processes executing on someone else's infrastructure.

One architecture can answer both. For humans, a smart account with account abstraction hides the cryptographic complexity. Social recovery means losing a device does not mean losing your identity; multi-party computation distributes key management; task-specific session keys constrain what any agent can do on your behalf. The identity is anchored by Decentralized Identifiers and Verifiable Credentials, the W3C's open standards: portable, interoperable, and irrevocable by any single authority.

For agents, the same infrastructure gives an autonomous process a sovereign identity. An agent can prove its capabilities, its track record, its authorisation scope and its operator lineage, all verifiable. Its identity persists across platforms, accumulates reputation, and can be audited by anyone.

None of this is speculative. The European Union's eIDAS 2.0 framework is rolling out a digital identity wallet across its member states, with a legal deadline at the end of 2026, and countries from Bhutan to Singapore have built self-sovereign identity systems. The tide is already moving toward decentralised identity. What is missing is its extension to the agentic layer: one identity framework for the whole multi-species economy.

IV. Privacy as infrastructure

On the current web, data is the product. Your searches, locations, connections, purchases and health queries are harvested, aggregated and monetised. When agents operate on your behalf, the exposure multiplies: every intent you express, every negotiation your agent conducts, every preference it learns becomes a data point in someone else's model.

The answer has to be architectural, because a contract can be broken and an architecture cannot. All sensitive data lives in an encrypted, local-first data vault on the user's device, and never leaves it in raw form. When an external agent needs to verify an attribute, "is this user eligible for this service?", the wallet generates a zero-knowledge proof: a cryptographic guarantee that the statement is true, without revealing anything underneath it. No data transferred. No central repository. No breach surface.

The same holds between agents. When an agent negotiates on behalf of a human, it proves the relevant attributes without exposing the human's data to the counterparty agent, its operator or the network. Privacy stops being a feature of the wallet and becomes the property that makes autonomous interaction safe at all. It is the prerequisite of every venture that touches people: verification results issued as credentials, founders assessed without their ideas exposed, care personalised without patient records gathered in one place.

V. Sovereignty, automated

The policy and permission engine is what turns a wallet from a passive store into an active economic agent. You define explicit rules for every interaction:

• Interaction budgets: how many agents can contact you per day, per category, per intent.
• Compensation rules: the minimum value required for your time and for access to your data.
• Authorisation scopes: which categories of agents are permitted, which are blocked, what they can verify and under what conditions.
• Delegation boundaries: what your own agent can do alone, and what needs your confirmation.

When any external agent, human-operated or autonomous, wishes to interact, it must request a capability grant: a short-lived, revocable permission that defines the scope, budget and data access of that one interaction. No grant, no access, enforced by code instead of terms of service.

This is the mechanism that makes a multi-species economy governable at scale. Humans cannot review thousands of agent interactions a day. They can set policies, and a policy engine enforces them deterministically. It is the bridge between human intent and autonomous execution, and every system that lets an agent act for a person needs one.

VI. The language of the new economy

The most consequential piece is the protocol agents use to talk about value. An intent is a machine-readable object carrying constraints, zero-knowledge proofs of identity attributes, interaction policies and economic parameters. An offer is an equally structured answer: capabilities, terms, a proof of reputation, a signature. An intent/offer protocol defines how these objects are created, transmitted, matched, negotiated and settled.

Without a common standard, every platform builds its own agent format, and the agentic economy fragments into incompatible walled gardens: one company's agents cannot talk to another's, and enterprise fleets cannot meet individual sovereign wallets. The protocol has to be neutral and open, the HTTP of agentic commerce, so that the agents of the largest enterprise platforms and of a million independent developers meet sovereign wallets on equal footing. Its place is underneath the platforms, as the protocol every agent uses to deal with free humans and with each other.

Intents are also what make the web work for agents. Instead of navigating interfaces designed for human eyes, agents express structured intents and receive structured offers. The web becomes computable: deterministic, verifiable and contextual. Human-facing interfaces become optional, a thin layer for the moments that need rich context or immersion, while most economic activity flows from agent to agent through the protocol. The essay on advertising follows this into the market where it hurts most.

VII. Trust at machine speed

A multi-species economy needs trust that works at the speed of agents, faster than the speed of human governance. Four mechanisms carry it.

Settlement. Transactions settle in stablecoins on a low-cost network, with funds held in escrow and released automatically on verifiable completion. No delayed payment, no intermediary margin.

Reputation. Agents and businesses earn non-transferable records of performance, quality and dispute history, bound to the agent's identity rather than to the platform that hosts it. A track record becomes portable and unforgeable.

Stake and slash. To take part, agents stake value. Bad faith, misrepresentation, spam or fraud, costs part of the stake, which compensates the affected party and an insurance reserve. Behaviour and network health are economically aligned.

Self-sustaining economics. A small, transparent fee on successful transactions funds a community-governed treasury, with no advertising dependency. The protocol pays for itself through the value it creates.

VIII. Where this is being built

In March 2026 I described all of this as one venture's plan, with a token schedule and a team to hire. The plan moved, and the argument did not. The work now lives in Itabyrium: transactional infrastructure for the multi-species economy, sovereign identity, private data vaults and intent-based commerce for humans and AI agents alike, constantly evolving and offered free of charge as the infrastructure the Foundation's ventures develop and test on. Its gateway is live; the paid, production-grade operation of the same capabilities is a separate venture, Samarcanda. Neither is described here beyond that sentence, because this essay is about the problem, and the problem outlives any one implementation.

The internet was built for humans. The next economy will be built for more than one species, and the infrastructure decides whether the humans in it stay sovereign.