The internet of representatives

>_GET /blog/internet-of-representatives

The web we know was built for human eyes. A page exists to catch attention, hold it, and bend it toward a decision: headlines, hero images, social proof, and a carefully placed button. Search, branding, copywriting, and "contact sales" are all instruments of the same craft: persuasion. For thirty years the open question was how do I convince the person reading this?

That question is quietly going obsolete. More and more, the thing on the other end of the connection is not a person but a representative, an agent acting on someone's behalf. It does not browse. It does not feel. It asks a colder set of questions, and persuasion is not on the list.

This post is the thesis for everything that follows. The internet is turning from an interface of persuasion into a layer of verifiable declarations between representatives of interests. Four shifts make the case.

1. The old internet was for humans

You might object here: we've been writing for robots for twenty years. Crawlers, meta tags, keyword density, the whole SEO game. That's true, but Google's crawler was just a matchmaker. Its only job was to drop your page in front of human eyes, and those eyes still had to be persuaded. SEO was just persuasion with a machine detour. The final reader, the one who actually signed the check, was always a person.

So the human web optimizes for attention. Its native artifacts, like landing pages, SEO rankings, brands, lead forms, and sales calls, are each a mechanism to win a person's trust through impression. None of them survives contact with a representative that reads structured data and discards the rest. To an agent, "contact sales" is not an invitation but a wall, and a salesperson's charm does not transfer, because trust has to travel as something checkable.

2. The new internet is for representatives

In fact, SEO is the precedent that proves the point. The moment a machine became the gatekeeper of attention, the entire web reorganized around what that machine could read. That same reflex is about to fire again. The only difference is that this time, the machine isn't the gatekeeper. It is the counterparty.

An agent acts in someone's name. Before it engages, it has to resolve four questions about the counterparty, and they are the same four whether the counterparty is a vendor, a clinic, or a person:

• Who are you? A real, identifiable entity behind a verifiable identity, in the sense of a W3C Decentralized Identifier.
• What can you do? Capabilities stated as machine-readable facts, closer to schema.org than to adjectives.
• Do you have a mandate? Authority to make the promise, which is a verifiable credential it can check.
• Are you telling the truth? Claims backed by evidence it can test, not assert.

You don't answer these questions with a nicer landing page. You answer them with signed, structured data built for software that cannot be charmed.

And nothing here is specific to software. The unit is not "buyer" or "seller." It is a representative of an interest meeting another. Change the principals and the same four questions reappear under a new name:

• Recruiter agent vs candidate agent. The personal page becomes an agent-readable profile: verified experience, projects, references, availability, constraints, and proof of work.
• Investor agent vs startup agent. The pitch deck becomes a manifest: traction, metrics, cap table, risks, proof points, and signed updates.
• Client agent vs service-provider agent. Freelancer, firm, or agency: reputation from delivered work, scope, deadlines, prices, and evidence of quality.
• Patient agent vs clinic agent. Availability, qualifications, outcomes, certifications, privacy, costs, and risk.
• Your agent vs a hundred app agents. The phone stops being a grid of icons. Every app exposes its capabilities, and "add cream to the shopping list" gets routed to the one that can do it. If none can, a marketplace is searched by capability, not by name.

And one pair is closest to home: a person's agent meeting a company's agent. A rough version already exists. A LinkedIn profile is structured claims about experience, skills, and references, except none of it is verified and it is read by human recruiters and keyword filters. Make the claims checkable and swap the reader for a recruiter's agent, a bank's agent, or a client's agent, and the personal site stops being a portfolio for human eyes. It becomes a verified personal surface: the interface through which other agents check who you are, what is confirmed, and what is merely claimed. The homepage, in other words, becomes a credential.

3. Trust becomes infrastructure

If persuasion no longer carries the decision, something has to. That something is a stack of verifiable trust, which is the subject of the rest of this series:

• Identity: who, provably (Are you really you?)
• Manifests: what, in machine-readable terms (The end of landing pages)
• Fit: whether it maps onto the need (Does this actually fit?)
• Pricing: cost as an API, not a negotiation (Pricing as an API)
• Sandboxes: proof on your own data (Measure, don't believe)
• Reputation: a record of kept promises, built from signed receipts (Reputation as capital)

Underneath the stack sits a quieter shift: for an agent, trust is not a sentiment but a computed value, a function of signatures verified, claims measured, and receipts accumulated. Sellers in this world stop optimizing for conversion and start optimizing what that function returns. The cultural shift comes down to one phrase: not "trust me", but "verify me."

4. The economy moves from attention to verification

When the decision rests on proof, the money follows. The cost of winning a customer stops being the cost of attention and becomes the cost of establishing trust. Marketing budgets migrate from ads and SEO toward proofs: identity, attestations, sandboxes, and signed track records. And reputation becomes capital. It amortizes the cost of every future transaction, because the next representative can check your record instead of taking your word. Even the funnel metric changes: what matters is no longer conversion but time-to-verified-trust, which is how fast a counterparty can reach a justified yes or no (Time-to-verified-trust). The whole economic flip is its own topic: The new economy of trust.

One company, two surfaces

To make the shift concrete, picture the same vendor as a human sees it and as a representative reads it. The human gets a sentence engineered to feel good:

<h1>Grow faster with the platform 5,000 teams trust 🚀</h1>
<p>Best-in-class automation. Loved by customers. Book a demo!</p>

The representative gets none of that. It fetches a signed object and computes:

{
  "did": "did:web:sam.example",
  "name": "Sam",
  "capabilities": ["workflow_automation", "data_export", "audit_reports"],
  "pricing": "https://sam.example/.well-known/pricing.json",
  "attestations": [
    { "type": "SOC2", "status": "verified", "issuer": "auditor.example" }
  ],
  "reputation": "proof-of-transaction | 5120 signed deals | 12m",
  "signature": "ed25519:9c1f...a47b"
}

Same company, two surfaces. The first is built to persuade; the second is built to be verified. The whole series is about what it takes to make the second one real.

Where we start: commerce

The pattern is general, but we need a concrete place to start. We follow one archetype in depth before gesturing at the rest. We start with the densest and most developed one: B2B software, where an agent buys a product on behalf of a company. Two recurring characters carry it. Blake is the buyer's agent (mnemonic: B = Buyer); Sam is the seller's agent (S = Seller). Everything in the trust stack will play out in the tension between them. Once you have seen it once, you will recognize it in every other pair above.