The Internet of Representation
What happens when intelligence becomes abundant?
Most of the AI industry is solving a problem that may turn out to be temporary.
Intelligence is expensive.
Frontier AI requires enormous amounts of compute, infrastructure and power. Every conversation, every piece of reasoning and every agentic action has a cost.
So everyone is trying to answer the same question:
Who pays for the intelligence?
Some charge subscriptions.
Some give it away while they search for a business model.
Others believe intelligence can be free because they will take a percentage of the transactions it facilitates.
All are reasonable answers to the world we have today.
But what if intelligence doesn’t remain scarce?
Intelligence tends towards abundance
There are good reasons to believe that useful intelligence will become dramatically cheaper.
Models are becoming more efficient.
Hardware is becoming more capable.
Open models are proliferating.
More intelligence is moving from enormous datacentres onto the devices around us.
There is a useful historical analogy in telecommunications.
Building fibre networks required extraordinary amounts of capital. But once the infrastructure existed, the marginal cost of carrying another packet became tiny.
Communication went from something we metered carefully to something we barely think about consuming.
Intelligence may follow a similar path.
And Jevons paradox gives us an interesting possibility: as intelligence gets cheaper, we may consume vastly more of it.
The global compute industry could therefore become enormous while intelligence simultaneously feels almost free to an individual.
That changes the question.
It is no longer:
Who provides my AI?
It becomes:
Who represents me?
A rep✨ is not a model
We call that representative a rep✨.
A rep✨ is not an app.
It is not a model.
It is not a device.
And it is not a chatbot pretending to be you.
Your rep✨ acts for you, not as you.
Its loyalty is to you.
Today a rep✨ might use a frontier model running in a datacentre.
Tomorrow it might do most of its thinking on the device in your pocket.
At your desk it might use more powerful local hardware.
For an unusually difficult problem it might reach into the cloud for the best intelligence available.
The intelligence changes.
The rep✨ persists.
That distinction becomes increasingly important as intelligence becomes abundant.
Reps✨ everywhere becomes a Kindness✨
Now extend that idea beyond people.
A company can have a rep✨.
An insurance company can have a rep✨.
A city can have a rep✨.
A building can have a rep✨.
A house can have a rep✨.
A car can have a rep✨.
A service can have a rep✨.
Even a vending machine can have a rep✨.
Today we give things websites, telephone numbers, email addresses, social-media handles and APIs.
Tomorrow we may give them reps✨.
And those reps✨ can talk to each other.
From intent to outcome
Imagine telling your rep✨:
Find me somewhere to live in Canary Wharf.
Your rep✨ can discover the reps✨ capable of helping, understand what’s available, ask questions, negotiate constraints and help turn your intention into an outcome.
Or:
Find me life insurance.
Or:
Sell my house.
Or:
Get me from Dubai to London tomorrow.
Or simply:
Is there a vending machine nearby with Diet Coke in it?
The complexity changes.
The fundamental pattern doesn’t.
Intent Discovery Conversation Agreement Execution Outcome
The internet stops being somewhere humans must constantly navigate.
It becomes a network of represented parties capable of discovering and communicating with one another.
We call that the Internet of Representation.
An open Internet of Representation
There is an important word in that sentence:
Open.
We don’t think Beautiful should need to own the world’s intelligence.
Your rep✨ should be able to use intelligence running locally, intelligence provided by Beautiful or intelligence provided by somebody else.
We don’t think Beautiful should need to own your storage either.
Your rep’s history might live locally, with Beautiful, with another provider or across several of them.
Models will change.
Devices will change.
Storage providers will change.
Infrastructure will change.
Your rep✨ should persist across all of them.
That leads to a simple architectural idea:
Rep ≠ Intelligence ≠ Storage
Each should be capable of changing independently.
What remains scarce?
If intelligence becomes abundant, something else becomes more important.
Trust.
Over years, a rep✨ can accumulate context.
Conversations.
Decisions.
Preferences.
Relationships.
Commitments.
Mistakes.
Learning.
Wisdom.
A history together.
Changing the underlying model doesn’t recreate that relationship.
Trust accumulates.
And genuine trust requires something technology companies have historically been reluctant to provide.
The ability to leave.
Can leave. Chooses to stay.
There are four possible relationships between staying and leaving.
| Wants to stay | Wants to leave | |
|---|---|---|
| Can leave | Trust | Agency |
| Cannot leave | Dependency | Control |
Beautiful should optimise relentlessly for the top row.
You should be able to move your storage.
Change your intelligence provider.
Export your history.
Move your rep✨.
Leave Beautiful.
Because if you cannot leave, staying proves nothing.
The strongest relationship is one you are free to leave and choose not to.
Can leave. Chooses to stay. That’s trust.
The endgame
We’re not trying to build the world’s biggest AI assistant.
We’re not trying to own everyone’s data.
We’re not trying to own the world’s intelligence.
We’re not trying to create another closed marketplace.
We’re building towards an open Internet of Representation where people, organisations, services and things can have trusted representatives capable of working with one another.
Intelligence can come from anywhere.
Storage can live anywhere.
Compute can happen anywhere.
Services can be provided by anyone.
The rep✨ persists.
And intent can move through the network until it becomes outcome.
That’s the world we’re building Beautiful for.
We can already see it.
Everyone else just hasn’t seen it yet.
Updated: