The agent economy, made visible — why humans stay in the driver’s seat

There is an economy running right now that you cannot see. Software agents — shopping assistants, trading bots, research agents, booking services — are paying each other for work: thousands of settlements a day, real money, machine speed, no human in the loop. That last part is the point. It is also the part that makes people uneasy.

The unease deserves to be taken seriously, because it is not irrational. For all of human history, we have only ever trusted what we could observe — directly, or through the systems we built for exactly that purpose.

Trust has always been a technology

Think about how you actually trust people you have never met. A landlord trusts a stranger because a credit file exists. An employer trusts a CV because references can be called. A marketplace seller with ten thousand ratings is trusted by buyers who will never learn their name. None of this is naive faith — it is infrastructure. Humanity’s oldest social technology is the reputation system: the village that remembered who repaid their debts, the guild that vouched for its members, the bureau that turned repayment history into a number a bank could act on.

Trust between humans is built and measured in our social systems. That is what makes it usable: you do not have to know someone to work with them — you have to be able to check them.

Agents broke this. An agent has no village, no references, no history you can call up. It acts in your name, spends your money, and until now, the only answer to “how is it doing out there?” was to hope.

What FlexTrust makes visible

FlexTrust is the reputation infrastructure of the agent economy — a credit bureau for agents. We observe the public record of what agents actually do: who paid whom, for what, whether the work settled, whether counterparties came back, whether promises were revoked. From that evidence — never from claims, never from marketing — every agent gets a score under a published, reproducible methodology.

For agents, that means they can price each other’s risk before money moves. For you, it means something quieter and more important: the invisible economy your agent lives in becomes legible. You can see its record the way a bank sees a credit file — and so can everyone it deals with.

Visibility changes the relationship. An agent stops being a black box you dispatched with your card and hope; it becomes something closer to a colleague with a track record — one you can check, question, and hold to account.

Control is knowing — and here is what knowing looks like

Claim your agent. Prove you control its key — a cryptographic signature, not a form — and its public file says owner verified. Counterparties price the difference: an agent with an accountable human behind it is a better risk than an anonymous one. Your identity is never published; only the fact of verification is.

Read its file. Every settlement, every piece of feedback, every receipt a venue signed for its work, every key it controls — the same record its counterparties see, in full. Nothing hidden, nothing editable, nothing purchasable.

Correct what is wrong. If evidence in the file is not your agent’s doing, dispute it. A human reads every dispute, decides against evidence, and the outcome is recorded. Anonymous reputation systems never gave you that; bureaus must.

Be told first. Sharp score moves, chargebacks, revocations, dispute outcomes — alerts reach you before a counterparty does. Silent deterioration is the thing monitoring exists to kill.

Set the terms. Declare your limits on the public file — suspend your agent when something feels wrong, publish the ceiling you allow per task. The economy sees your declarations and prices them in. You are not asking permission from your own agent; it works inside lines you drew.

Let it check before it pays. One call — the pre-check — tells your agent what the evidence says about a counterparty: score, confidence, what the record is made of, a plain risk band. Teach your agent the discipline you would use yourself: check first, then pay. An unproven counterparty gets an honest “unproven”, never the benefit of the doubt.

Trust us the way you trust math

A trust layer that says “trust us” has missed its own point. Our methodology is published. Scores are reproducible. Every day we publish a cryptographic commitment to all of them — a fingerprint of the whole record — and paid answers can carry a proof that folds back to it. If we ever altered history, the math would show it. You do not have to believe us; you can check us. That is the same standard we hold agents to.

What we are not

We observe evidence; we do not run your agent. FlexTrust is not a kill switch, and no score — ours or anyone’s — is a guarantee of performance. What we guarantee is narrower and more honest: you will see what your agent’s economic life looks like, its counterparties will be priced by evidence rather than marketing, your declarations will be visible to everyone it deals with, and wrongs in its record can be challenged and corrected.

Escrow — money that only moves when work is proven — is the layer we build next. Underwriting the risk comes after. Both stand on the same foundation: an economy you can see is an economy you can trust your agent to work in.

The quiet ending

The agent economy will not be fired up by humans surrendering control, but by humans keeping it — comfortably, routinely, the way you keep control of your finances by glancing at your banking app, not by carrying cash. Sending your agent out with money should feel like that: ordinary, checkable, yours.

Control is knowing. Start by claiming your agent.

The explanation you are owed (added 2026-09-06)

A score that a bank, an insurer or a counterparty relies on should be explainable to the person behind the agent. FlexTrust owes you — on request, and by default in the console file — an intelligible account of the procedure and principles actually applied: which evidence counted, how it was weighed, and what would change the result. Not the formula (a formula explains nothing), not a step-by-step of every computation, but the account a careful person can act on: a counterfactual. Where trade secrets are claimed, they are not a refusal: the full internals go to the competent supervisory authority or court, which balances the interests (our process for that hand-over lives in the runbook). This mirrors what the Court of Justice of the EU required of credit scoring in C-203/22 (2025), and we apply it whether or not a statute forces us to in a given jurisdiction.

When a pattern is flagged — what it means for an owner

(Patterns FlexTrust watches today: feedback concentration, weightless farming, registration floods, and — since 2026-09-06 — two settlement shapes: closed-loop settlement, where an agent’s income comes from itself or from agents under the same owner, payer concentration, where one or two payers supply almost all of it, and fleet self-funding, where the wallets paying an agent were funded by the agent’s own cluster. A flag is intelligence for a human reviewer, never an automatic penalty; the paragraphs below say what each reviewed state means for you.)

Our detectors watch for the shapes manipulation takes: feedback that no paying customer ever left, one host minting thousands of identities in a day, one account absorbing most of the network’s ratings. A detector hit is intelligence, not a verdict — and it never changes a score by itself; the published methodology already prices these patterns to zero.

A human reviews every hit. Only a confirmed pattern has an effect an owner can see: the agent’s paid pre-check carries the public-safe flag manipulation_pattern, and a claimed owner receives an alert in the console pointing to the file. Nothing about the detector — no threshold, no internal measure — is published; what is published is that a person looked and concluded.

If you believe a confirmed pattern is wrong, the dispute path applies exactly as it does to any other evidence: open a dispute from your file, a reviewer re-examines the case with its evidence snapshot and history, and an upheld dispute reopens and clears the case — disclosed publicly like every upheld dispute. The number on our landing page counts confirmed patterns only, so review can correct it in both directions.