OpenAI, Visa and Mastercard push AI agents into commerce as trust becomes part of the transaction

AI · COMMERCE · TRUST · REPUTATION

OpenAI, Visa and Mastercard push AI agents into commerce as trust becomes part of the transaction

AI agents are beginning to shop, book services, buy compute and make payments on behalf of people and companies. But machines do not value pounds, dollars, dinar or rupees in the human sense. Money matters to them only because it unlocks something else: access, compute, data, permissions, time and reliable counterparties. As autonomous systems become economic actors, reputation could emerge as one of the most important currencies machines use to decide who — and what — they trust.

Humans have spent thousands of years deciding what counts as money.

Gold.

Silver.

Cowrie shells.

Pounds.

Dollars.

Dinar.

Rupees.

Numbers stored inside bank databases.

The objects changed.

The human reason for wanting them did not.

Money could be exchanged for food, shelter, status, security, freedom, labour and almost anything else somebody desired.

Artificial intelligence introduces something unfamiliar into that arrangement.

A machine does not need dinner.

It does not need a bigger house.

It does not dream about retirement.

It does not admire the colour of gold.

It does not become wealthier in any meaningful psychological sense when its bank balance increases.

So what happens when we have to bargain with an economic actor that does not value money the way humans do?

That question is beginning to leave philosophy

AI agents are already moving from giving advice to taking actions.

OpenAI's Agentic Commerce Protocol allows AI systems, merchants and people to work together to complete purchases.

ChatGPT can already help users move from product discovery toward checkout while merchants continue to handle payments, fulfilment and customer relationships.

Visa says AI agents are beginning to book travel, reorder inventory, purchase data and buy computing resources on behalf of businesses and individuals.

Mastercard has gone further and built infrastructure specifically for machines to transact with other machines.

Its Agent Pay for Machines system is designed for programmatic, always-on transactions that can happen at machine speed, including payments worth only fractions of a cent.

The machine economy is therefore beginning to take physical shape.

But the machine does not want the money

This distinction matters.

When a human earns £1,000, the money itself carries broad optionality.

It can be saved.

Spent.

Given away.

Invested.

Used to improve comfort or status.

An AI agent has no equivalent human desire.

Money is useful to the system only insofar as it enables an objective.

It may allow the agent to buy:

• Compute
• Cloud storage
• Specialist data
• API calls
• Bandwidth
• Software tools
• Access to proprietary systems
• Human labour
• Priority processing
• Physical goods required for a task

Remove those possibilities and the currency itself has little meaning to the machine.

For an AI, money may be less a possession than a permission slip.

That changes the meaning of wealth

Imagine two autonomous systems.

One controls £1 million but has no permission to purchase compute.

The other has access to a large computing cluster, premium datasets and trusted APIs but almost no cash.

Which one is wealthier?

For a human, the answer initially seems obvious.

For a machine, it may not be.

The second system may possess far more useful capability.

Machines may value access more than ownership

Human economies developed around scarcity.

AI economies will too.

But the scarce things may look different.

High-quality data can be scarce.

GPU time can be scarce.

Low-latency network access can be scarce.

Permission to call a proprietary API can be scarce.

Access to authenticated humans can be scarce.

Reliable information can be scarce.

Trust can be scarce.

Those forms of scarcity may become more economically meaningful to an autonomous agent than possession of any particular currency.

So how would humans barter with AI?

The answer may be surprisingly old-fashioned.

We would exchange something the other side needs.

A machine might provide analysis in return for access to a dataset.

One agent might provide computing capacity in return for specialised model access.

A company might give an agent permission to use an internal system in exchange for completing a task.

An AI network could allocate processing resources according to the reliability of other participating systems.

Money may still sit underneath many of these exchanges.

But it would increasingly become settlement infrastructure rather than the thing the machine ultimately values.

The most valuable asset may be certainty

Consider an autonomous purchasing agent trying to acquire 20,000 parts for a factory.

It finds two suppliers.

Supplier A is 8% cheaper.

Supplier B has delivered 4,000 previous orders on time, has verified ownership, authenticated inventory, transparent dispute records and a strong history of resolving problems.

A human buyer may describe Supplier B as trustworthy.

The machine may describe the same information differently.

Lower expected failure probability.

Lower verification cost.

Lower fraud risk.

Lower likelihood of fulfilment failure.

Lower expected dispute cost.

Different language.

Same economic value.

Reputation may become valuable to machines precisely because it compresses uncertainty.

That could make reputation function like currency

Currency does several important things.

It communicates value.

It reduces the complexity of exchange.

It allows strangers to transact without negotiating every underlying resource individually.

Reputation can perform a similar function in a world of autonomous agents.

A strong reputation record could tell a machine:

• This identity has been verified
• This counterparty has completed previous exchanges
• Other participants corroborate its history
• Disputes have occurred
• Responses were recorded
• Some disputes were resolved
• The record has remained consistent over time

The result is not money.

But it can alter access to economic opportunity in much the same way money does.

A machine may pay more for someone it trusts

Humans already do this.

We pay established companies more than unknown ones.

Banks offer different borrowing terms according to risk.

Insurers charge different premiums.

Employers pay experienced people more because past performance reduces uncertainty about future performance.

AI agents could make the same trade-off continuously and mathematically.

A machine might willingly pay 3% more to a supplier whose verified record reduces expected fulfilment risk by 10%.

Reputation would have acquired a measurable economic value.

Trust is already moving into payment infrastructure

Payment companies are explicitly confronting this problem.

Visa has warned that agentic payments introduce another layer between the human authorising a transaction and the financial institution processing it.

That means traditional questions such as identity are no longer enough.

Systems increasingly need to understand authority, intent and context.

Mastercard has similarly argued that machine commerce will require trusted, secure and reliable mechanisms because agents may transact continuously without a person manually initiating every payment.

In other words, the future payments problem is not merely:

Does this account contain enough money?

It is also:

Is this machine authorised, is this transaction legitimate and can the counterparty be trusted?

Identity alone will not solve that

Identity establishes who or what something is.

It does not establish how it behaves.

A company can be perfectly real and still fail to deliver.

A person can pass identity verification and still break an agreement.

An AI agent can be cryptographically authenticated and still repeatedly make poor decisions.

The machine economy will therefore need both identity and history.

An AI may need a Reputation Record of its own

Imagine an autonomous procurement agent operating for five years.

It negotiates thousands of purchases.

It chooses suppliers.

It settles invoices.

It disputes incorrect deliveries.

It makes mistakes.

It improves.

Eventually another company is asked to allow that agent access to its systems.

What should the company examine?

The model name?

The company that built it?

The latest benchmark score?

Or the history of what that particular agent actually did?

When machines become actors, machine reputation may matter as much as machine intelligence.

The same applies when AI evaluates humans

The reverse relationship may be even more important.

Autonomous agents will increasingly make recommendations about people.

Contractors.

Tenants.

Employees.

Suppliers.

Borrowers.

Business partners.

If the machine receives only a name and a payment method, its view is narrow.

If it receives verified history, evidence, responses and outcomes, its view becomes richer.

That creates both opportunity and serious governance questions.

Reputation cannot become an invisible machine score

There is an obvious danger.

A reputation economy run entirely by machines could become deeply unfair if people cannot understand, contest or correct the information being used against them.

A low machine-generated score should not quietly become a permanent economic sentence.

People need to know:

• What information was used
• Where it came from
• Whether it was verified
• Whether conflicting evidence exists
• Whether an allegation remains unresolved
• Whether the person responded
• Whether the matter was later corrected or resolved

Reputation is useful only when the underlying record can be interrogated.

This is where RRSource's argument becomes relevant

RRSource is built around the idea that trust should be based on more than identity, reviews or a single snapshot score.

A Reputation Record can instead preserve the chronology of an exchange.

Claim.

Evidence.

Corroboration.

Response.

Resolution.

If machines increasingly make trust decisions, that structure could become valuable because it offers something AI systems need:

structured evidence about what happened before.

Machines may value provenance more than persuasion

Humans can be influenced by presentation.

Branding.

Confidence.

Appearance.

Prestige.

Familiarity.

An AI system may still inherit human biases, but it also has the potential to ask more mechanical questions.

Where did this information come from?

Was the source authenticated?

Is there independent corroboration?

Has the evidence been altered?

Did the other party respond?

What happened afterward?

Provenance can become more valuable than presentation.

That could change what people optimise for

Today's online economy often rewards visibility.

Followers.

Reviews.

Search ranking.

Advertising.

Brand recognition.

A machine-mediated economy may reward something different:

verifiable reliability.

If an autonomous agent can examine thousands of records in milliseconds, persuasive branding may matter less than documented outcomes.

A small supplier with an exceptional verified history could become more attractive than a famous company with a weaker operational record.

Reputation could become machine-readable capital

Financial capital gives its holder options.

Social capital gives its holder relationships.

Reputation capital gives its holder credibility.

In an agentic economy, that credibility could become machine-readable.

An AI system could price risk differently based on it.

Grant greater permissions.

Require smaller deposits.

Approve faster.

Offer better terms.

Or refuse a transaction altogether.

That is economically significant.

But reputation is not literally currency

The distinction matters.

Reputation is not fungible in the way money is.

Ten units of trust cannot simply be transferred from one person to another.

Reputation is contextual.

A brilliant surgeon does not automatically make a trustworthy building contractor.

A dependable tenant does not automatically become a good investment adviser.

Machine systems therefore need contextual reputation rather than a universal social score.

The value lies in the record, not merely the number attached to it.

Money will not disappear

Machines will continue using pounds, dollars, dinar, rupees, stablecoins and other payment instruments because humans, companies and governments organise economic life around them.

AI agents will require ways to settle obligations.

Payments networks will still matter.

Banks will still matter.

Pricing will still matter.

What changes is what sits immediately before the payment.

Which supplier does the agent select?

Which offer does it trust?

Which counterparty deserves access?

Which claim does it believe?

Which risk does it accept?

The answers increasingly depend on information.

The machine economy may become an economy of permissions

The most valuable question may eventually stop being:

How much money do you have?

And become:

What are you trusted to do?

Can the agent spend £100?

£10,000?

Can it negotiate contracts?

Access confidential data?

Hire another agent?

Purchase compute?

Commit the company legally?

Every one of those is a question of authority and trust.

Then reputation becomes leverage

A machine may not care whether somebody is famous.

It may care that they have delivered 99.7% of verified commitments.

It may not care about an expensive advertisement.

It may care that 200 previous exchanges contain authenticated evidence and resolved outcomes.

It may not care about a beautifully written testimonial.

It may care whether the testimonial came from a verified participant in a real transaction.

That turns reputation from a marketing asset into operational leverage.

Humans invented money because barter was inefficient

AI may introduce the next layer.

Money answers:

What can this party pay?

Reputation answers:

What can this party be trusted to do?

Autonomous commerce will probably require both.

Money settles the transaction.

Reputation may determine whether the transaction happens at all.

AI may not value dollars, dinar or rupees

It will nevertheless operate inside economies built around them.

The important change is that machines may judge value differently from the humans who created those economies.

Compute can matter more than cash.

Access can matter more than ownership.

Verified data can matter more than persuasion.

Reliability can matter more than reputation in the traditional branding sense.

And a documented history of successful exchanges can become a form of capital that machines understand immediately.

Humans built money to measure value. Machines may increasingly measure value through access, certainty and trust.

If that happens, reputation will no longer sit quietly beside the economy.

It may become part of the infrastructure through which the economy operates.

Editorial note: Current AI systems are not known to possess human desires for wealth, status or ownership. References in this article to what machines “value” describe the resources, permissions and outcomes that can become instrumentally useful to systems pursuing assigned objectives. Reputation functioning as a form of machine-readable economic capital is a forward-looking proposition, not an established replacement for money.