The Internet of Representation

The central question of the first Internet was: where do I find the information?

The central question of the next one is: who is acting, for whom, with what authority and under whose control?

Today that answer is scattered across accounts, passwords, platforms, terms of service, proprietary databases and private APIs. Every company creates its own version of you. Your identity lives in thousands of systems you do not control, your history is trapped inside products, your permissions vanish when you leave and your reputation does not travel. Every organisation knows something about you and none of it adds up to a way of representing yourself across them.

AI makes that failure impossible to ignore.

An AI system may know what you want without having permission to act. It may understand your preferences without being able to prove it represents you. The moment it sends a message, signs an agreement, makes a payment or accepts an offer, the receiving party needs to know six things:

The Internet has protocols for moving data, publishing documents, sending messages and transferring money. It has no shared, general-purpose protocol for representation.

That gap is not a new observation. Tim Berners-Lee set out a roadmap for a Semantic Web of connected applications in 1998. ProjectVRM began in 2006 to equip customers with tools of their own and later described "fourth parties" working for customers as third parties work for vendors. Today, an active field is building delegated authority for AI systems using decentralised identifiers and verifiable credentials. The idea does not need inventing. What changed is that it stopped being theoretical.

A representative should be able to say "I represent this person" and the recipient should be able to check that independently. Not because a platform claims it is true. Not because Beautiful sits in the middle. Because the authority can be proven.

That representative may be a person, an organisation, an AI or another delegated representative. Its authority may be broad or narrow, may last for years or for a single transaction and may be trusted for one purpose while untrusted for another. Representation is not a chatbot feature. It is the relationship between an identity and whatever acts for it.

Persistent representation used to be something only money could buy. A solicitor, a broker, an agent: someone whose job was to hold your interests inside a transaction and who charged accordingly. For the first time everyone can have that, not because we all became richer but because the cost of representation collapsed.

So there will not be one universal AI that does everything. People will have teams: one for work, one for renting, one for travel, one for health, one for money, one for the family. Some personal, some professional, some standing for years and some created for a single afternoon. They will collaborate, delegate, consult and negotiate with each other, always inside authority granted by the person they serve.

The person remains the principal. The representatives remain accountable.

That is the inverse of the platform model. Today a person enters a company's system and operates inside its rules. Here they arrive carrying their own identity, preferences, history and representatives. The centre of gravity moves from the platform to the person.

The same architecture runs the other way. Companies, institutions and public bodies will have representatives that answer questions, provide evidence, negotiate terms, expose services and take authorised action. A renter's representative talks to a building's representative. A candidate's talks to an employer's. A patient's talks to a healthcare provider's.

Transactions that today need forms, call centres, portals and three weeks of email become an exchange between two representatives that can each prove who they act for. The interface between organisations stops being a webpage.

A page presents information for a person to interpret and act on. A representative participates: it holds context, asks questions, makes decisions, negotiates and maintains a relationship over time.

Websites do not disappear but they stop being the primary operating surface. Search does not disappear but it becomes subordinate to intent. Apps do not disappear but they become one interface onto a wider network. The Internet stops being somewhere people visit and becomes an environment in which authorised software works on their behalf.

That only holds if the data underneath changes shape with it. Representation cannot run on unstructured pages, inaccessible databases and proprietary silos. A home should publish what it is, where it is, what it costs and whether it is available. A job should publish its requirements, conditions and status. An organisation should publish facts about itself in a form software can read and check.

Underneath, the model stays deliberately small. A Subject is anything the network can describe: a person, an organisation, a representative, a building or a car. A Connection is a typed link between two subjects. A Claim is a signed assertion. Evidence allows others to assess it. Four things. The discipline is to keep finding fewer rather than adding more.

Today digital trust hides inside brands and platforms. A person trusts a bank because it is a bank. A business trusts a platform because it controls the account. A service trusts a person because they know a password. That is indirect, weak and completely unsuited to software that acts.

The Internet of Representation makes trust inspectable instead. A claim shows who made it. An action shows who authorised it. A delegation shows its scope. A decision shows the evidence behind it. A representative shows the limits of its own authority.

None of that produces a score. There is no rating out of five for a person and no central bureau deciding who is reputable, because the moment a protocol issues a number it has quietly become the judge. What it publishes is the evidence. Trust does not become absolute. It becomes legible and two parties applying different policies can reach different conclusions from the same evidence. An open system needs exactly that.

A valid signature proves that a document was signed by whoever controlled a particular key. It does not prove that the statement is still current.

Every honest system has to answer that difference, because a stolen key can keep producing valid signatures and withdrawn authority can appear valid forever if nobody checks whether it still stands.

A recipient therefore needs both the signed document and reliable evidence of its present status. In Beautiful's design, independent witnesses provide that evidence, with the strength and freshness required scaling to what is at stake.

This is the hardest part of the design. We would rather expose it honestly than ship a green tick that means less than it appears to.

Any of this can be checked rather than believed. Beautiful publishes a machine-readable front door giving the identity method, the addressing scheme, a signed example document and a three-step verification recipe, with no directory and no Beautiful code involved.

A representative is only legitimate while the person it serves stays in control. Authority has to be deliberately granted, limited in scope, inspectable by the parties who need to check it, revocable, recoverable and auditable.

A representative should not own the relationship. A platform should not own the identity. A provider should not be able to trap a person inside its system. People have to be able to replace a representative without losing themselves, move provider without rebuilding their digital life, withdraw authority and recover from loss, compromise and failure.

The representative works for the person. The person does not belong to the representative.

The first generation of the Internet created open protocols. The second created closed platforms. The third has to restore openness at the layer that now matters most: identity, authority, representation and trust.

No single company should own that layer. It should be as open as publishing a website or sending an email. Anyone should be able to create an identity, build a representative, delegate authority, publish open data, verify a claim, run an index, build an application and implement the protocols, without asking permission from us or from anyone else.

A representative owes its duty to one party: the person who appointed it. Funding does not create that duty and it must never be able to bend it.

Someone still has to pay and it will usually not be the person served. The workable pattern is the one every market with agents already uses, where the counterparty pays for a good outcome. That arrangement is also the sharpest risk in the whole design, because money arriving from the other side is exactly how loyalty erodes. So the duty is fixed first and the payment is fitted around it, never the other way round. Who pays is published, what that payment can buy is capped and the representative is expected to cost its own operator money whenever the person's interest requires it. One that never once does has not proved anything.

Beautiful is building these foundations in public. Not to control the network. To make it possible. We expect to earn a living from the tools, the hosting and the representatives we run well, never from being the only door. The test is deliberately uncomfortable for us: the system only works if a person can leave Beautiful without losing their identity, authority, history, relationships or ability to appoint new representatives.

Platforms will not disappear but they should stop being the sole source of truth about who someone is, what they want or who may act for them. The most important digital relationships should not depend on one company continuing to exist.

Rentiful represents renters. Others will represent workers, travellers, patients, investors, citizens and organisations. Those products matter because they solve real problems for real people and they pay for the work. They are a proof of representation, not the boundary of the company.

Each one has to demonstrate the same thing: that a representative can understand a person, act inside delegated authority, talk to organisations, use open data and stay accountable to the individual it serves. The vertical changes. The architecture does not.

Today an organisation controls the customer relationship because it controls the interface, the data and the transaction. Representation gives the individual their own side of it. A buyer has a representative. A renter, a patient and a worker each arrive with something acting for them that is as capable as the software across the table.

For the first time both parties to a transaction can afford to be professionally represented. That reduces information asymmetry, lowers the cost of transacting, makes switching easier and exposes poor service faster. Competition moves from capturing attention to producing outcomes.

The deeper shift is what happens to the data. For twenty years the most valuable asset online has been the pile of information a platform holds about a person: the profile, the history, the relationships, the standing. That pile is the moat. Make it portable and the moat drains. Any business whose defence rests on holding data about the person rather than on serving them better than anyone else is now building on sand.

The dominant business model of the first Internet was capturing people. The next one can be built on serving them.