Alpesh Patel, Strategic Partnership Director at Cartex – Interview Series

Alpesh Patel is Strategic Partnership Director at Cartex, bringing more than 25 years of experience across fintech, cryptocurrency, card issuing, and payments in the UK and international markets. His career spans payments, remittances, and financial infrastructure, with an emphasis on how services integrate, perform for merchants, and handle operational risk. At Cartex, he helps bring payment services and partners together to simplify the complexity businesses face when moving money.

Cartex is a fintech marketplace and infrastructure platform bringing together payment gateways, card acquiring, payouts, alternative payment methods, card issuing, and digital asset services. It works with merchants, banks, wallet providers, and other online businesses to connect financial services through a shared platform. In this interview, Patel discusses agentic payments, delegated authority, consumer recourse, existing payment rails, and the signals that distinguish sustainable demand from experimentation.

You are helping build Cartex after spending much of your career in payments, remittances, and financial infrastructure. What led you to help build this company, and how has being an operator shaped the way you evaluate emerging fintech trends?

I have spent over 25 years in fintech and payments, and everywhere I looked, payment infrastructure was becoming increasingly complex. More layers were being built between the business and its money, and regulation grew stricter at almost every level. Businesses still want to accept and send money across countries and through different payment methods, but each new connection adds another integration and another compliance requirement.

Co-founding Cartex was the opportunity to streamline that complexity by bringing payment services and partners together. Working as an operator has made me practical: when I look at a new trend, I ask what has to work operationally. The real test is how a service performs for the merchant, how easily it integrates and who takes responsibility when a payment fails.

Agentic payments were one of the biggest fintech narratives entering 2026, yet transaction volumes have reportedly started declining even as hundreds of startups enter the market. Do you see this as a temporary correction, or are we beginning to discover that the actual demand for autonomous payments is narrower than the industry initially expected?

I think it is a bit of both. Reported swings in activity on x402, a protocol for internet-native payments, show why raw transaction counts need careful interpretation. They can include testing, automated activity, and self-payments, so they don’t necessarily reflect recurring customer demand. I would be cautious about treating one protocol as a measure of the whole agentic payments market.

So I wouldn’t take that as proof that demand has disappeared. A customer may be comfortable letting an agent compare prices or build a basket, but more cautious about letting it pay without asking. The industry is still working on the safety and permissions needed to build that confidence.

The current environment has been described as a potential “shakeout phase.” What typically happens during this stage of a fintech cycle, and what separates companies that survive the transition from hype to sustainable adoption from those that disappear?

Capital has become more selective, and companies that raised funds on a promising idea now have to show revenue to raise the next round. Some are acquired for their team or technology; others simply close outright.

Many people say we are approaching a stage of AI disillusionment, but I still see companies delivering real value to clients. Often, those are businesses working on something that rarely makes media headlines, like automating expense reports. Capital One’s $5.15 billion acquisition of Brex is one example of the value placed on combining corporate cards, expense management, and banking in a single platform.

At the same time, startups that offer something a large payment company can quickly replicate may struggle to sustain a standalone business.

Much of the conversation around agentic commerce focuses on what an AI agent can buy or transact, but the harder problem may be establishing who authorized the agent and what it is permitted to do. What kind of identity, permissioning, spending limits, and authentication infrastructure needs to exist before businesses can safely give AI agents control over money?

The industry’s main direction now is to prove who built the agent and to understand on whose behalf it acts. Banks and payment companies should know this before an agent can touch the money. That requires a verifiable credential linking the agent to its provider and to a customer who has already passed KYC checks. It also requires a signed mandate with spending limits and permitted categories, which the customer can revoke instantly.

Authentication needs to account for delegated authority. An agent cannot complete a biometric check the same way a person holding a phone can. The customer should authenticate when granting permission, and each transaction should be checked against that permission, with further customer approval where required. Delegation must not become a way to bypass the controls that protect the account.

Payments have traditionally relied on clearly identifiable consumers, merchants, banks, and payment processors. When an autonomous agent becomes another participant in that transaction chain, who should ultimately be responsible when it makes an incorrect, fraudulent, or unauthorized payment?

As I see it, responsibility should follow control. Every agentic transaction involves customers, developers, merchants and payment companies, and each side controls a different part of the risk. Each should be accountable for the controls it manages. The legal allocation depends on the applicable rules and agreements, but a clear process must exist to investigate and resolve a disputed payment.

If the agent stayed within a valid permission and the customer simply regrets the purchase, that is one thing, but if it was compromised or manipulated, for example, by hidden instructions on a web page, the question is whose controls failed and who should be responsible.

One challenge with autonomous payments is reversibility. Existing card networks have established mechanisms for refunds, disputes, and chargebacks, while some digital-asset transactions can be effectively irreversible. How do you think agentic payment systems should balance automation and instant settlement with the need for consumer protection and recourse?

I would separate settlement speed from consumer protection. Money can move instantly, but customers still need a way to challenge an unauthorised payment or resolve a failed purchase. For consumer purchases, card-style protections will remain important, and dispute frameworks will need to clarify how payments outside an agent’s mandate are handled.

However, many digital-asset transfers are irreversible, so protection has to be built in at a higher level. Mandates for AI agents may become narrower, or payment companies could restrict new recipients and require human confirmation if a transaction’s value exceeds an agreed limit.

The real risk is a system that combines full automation and instant finality without understandable accountability when something in a payment goes wrong.

Through Cartex, you work with infrastructure spanning acquiring, payouts, alternative payment methods, cards, and digital assets. From that infrastructure perspective, will agentic commerce require entirely new payment rails, or can most of it ultimately be built as an intelligence and permissioning layer on top of existing financial networks?

We mostly work as a layer on top of existing networks. Cards, account-to-account payments, wallets and payout rails already handle settlement and fraud screening. In most cases, an agent needs a reliable way to prove who it is and what it can spend, not a new way to transfer money.

One area where different economics may be needed is high-frequency payments between machines, such as paying per API call. Processing each fraction-of-a-cent payment as a separate card transaction is difficult to justify. Stablecoin-based rails may be useful here, provided network fees, conversion costs and compliance requirements make sense. For most merchant purchases, though, I expect agents to use existing infrastructure, with additional controls to verify their authority.

With so many startups now positioning themselves around agentic payments, what metrics or signals do you look for to distinguish genuine transaction activity and customer demand from pilots, demonstrations, or AI features added primarily to attract investor attention?

When assessing a potential payments partner, I first want to understand the customer problem it solves. Then I look at the quality of volume, how many merchants transact repeatedly, and how much volume comes from the top few counterparties. Automated activity alone doesn’t prove demand, so you need to know who uses the service and why. I also ask about net revenue, take rate, fraud and dispute rates, and cohort retention, since they show whether merchants who joined a few months ago still use the product at the same or higher volume.

Early-stage companies may not yet have enough history to show all of this, so I look at signed partner agreements, the scope of live pilots and regulatory readiness. From a partnership perspective, one of the most important questions is still if the team can clearly explain what happens when a payment goes wrong.

Looking three to five years ahead, what do you expect the agentic payments market to actually look like after the current experimentation and consolidation plays out, and which parts of the financial stack—identity, compliance, orchestration, settlement, fraud prevention, or something else—do you believe will capture the most durable value?

I expect agentic payments to become one of the channels, much as mobile payments did. Much of the volume will probably still run through established payment providers and wallets that have already built networks of consumers and partners. As for startups, I believe many will be acquired, shift their focus or shut down.

The real value may be in the systems that can reconstruct what an agent was authorised to do and what actually happened. Many people are still cautious about letting AI handle their money, so companies need clear limits, records, and processes to prevent and resolve payment errors.

That requires a clear record of which agent initiated the payment, for whom and within what limits. A provider that connects those records to transaction monitoring, reconciliation and dispute handling can become an important operational partner. Its value comes from helping banks and merchants manage agent payments reliably as volumes grow.

Thank you for the great interview, readers who wish to explore agentic payments, payment infrastructure, and the controls behind delegated financial transactions should visit Cartex.