Visa's Intelligent Commerce initiative and PayPal's agentic payments push have done something useful: they've moved the agent-to-agent commerce conversation from speculative to structural. Both announcements signal that the largest players in global payments believe autonomous AI agents will soon be initiating, negotiating, and completing purchases on behalf of individuals and organisations — without a human clicking 'confirm' at any point in the chain. The question for UK fintechs and enterprise technology leads is no longer whether this is coming. It is already arriving. The question is which part of the existing infrastructure breaks first.
The answer, increasingly, is settlement finality. AI agents can operate at millisecond speed across complex procurement decisions. The payment rails they're expected to use were designed for humans — with confirmation loops, fraud review windows, and chargeback rights that presuppose a person was involved in initiating the transaction. Strip out the human, and you expose a gap that no amount of clever API design fully papers over. This is the defining infrastructure problem for the next wave of commerce, and UK fintechs are better positioned than most to address it — if they move deliberately.
What Agent-to-Agent Commerce Actually Looks Like
It helps to be precise about the use case, because 'agentic payments' gets used loosely. The scenario that creates genuine infrastructure stress is not an AI assistant reminding you to renew a subscription. It is an autonomous procurement agent — operating under a defined policy mandate from a CFO or operations director — identifying a supplier, negotiating terms, committing to a contract, and triggering payment, all without synchronous human approval. The human set the parameters; the agent executes within them. This is already happening in limited forms in areas like programmatic advertising, automated inventory replenishment, and algorithmic trading adjacent to supply chain finance.
The commercial logic is compelling. For organisations managing high-frequency, low-value procurement — consumables, software licences, data feeds, logistics spot-pricing — removing human confirmation loops from each individual transaction dramatically reduces friction and cost. Visa's framework explicitly anticipates agents holding payment credentials and acting on delegated authority. PayPal's approach focuses on credentialled agent identities that can be granted scoped spending permissions. The architecture is coherent. The gap is in what happens after the agent commits.
The Settlement Finality Problem in Plain Terms
Settlement finality refers to the point at which a payment is irrevocable — when the receiving party can be certain the funds are theirs and the transaction cannot be unwound. In traditional payment rails, finality is deliberately delayed. Card networks build in chargeback windows of up to 120 days precisely because human error, fraud, and dispute are expected features of the system. Faster Payments in the UK achieves near-real-time clearing, but the dispute and indemnity frameworks still assume human-initiated transactions. When an AI agent commits to a purchase that is later disputed — because the agent misinterpreted a policy parameter, encountered a vendor acting in bad faith, or executed under a compromised credential — existing frameworks struggle to assign liability cleanly.
This is not an abstract risk. It is a material barrier to enterprise adoption. A head of finance authorising an autonomous procurement agent needs to know, with legal and operational certainty, what happens when the agent makes an error. Who bears the loss? Over what window can the transaction be reviewed? What audit trail satisfies both internal governance and external regulatory scrutiny? At present, the answers depend heavily on contractual workarounds and manual exception processes — which rather defeats the purpose of autonomous operation. The infrastructure layer needs to catch up with the agent capability layer, and it has not done so yet.
Where the UK Fintech Ecosystem Has an Advantage
The UK is unusually well-placed to lead on this problem. The combination of the Payment Systems Regulator's active posture, the FCA's sandbox infrastructure, and the maturity of the open banking ecosystem gives domestic fintechs both the regulatory engagement pathways and the technical primitives to prototype solutions that other markets lack. The New Payments Architecture programme, though its rollout has been protracted, is specifically designed to modernise the underlying settlement infrastructure in ways that could accommodate non-human counterparties if the policy intent is directed there.
More concretely, several technical approaches are under active development. Programmable payment logic — whether implemented via smart contract mechanisms on regulated rails or via rule-sets attached to agent credentials — can encode the conditions under which an agent transaction achieves finality automatically, reducing the need for after-the-fact dispute resolution. Cryptographic audit trails linked to agent identity frameworks (analogous to the way open banking uses OAuth-scoped consent) can provide the governance evidence that enterprise finance teams require. The building blocks exist. What is missing is a coordinated effort to assemble them into a coherent settlement model specifically designed for non-human counterparties.
Liability Frameworks Cannot Be an Afterthought
Technical infrastructure alone will not resolve the adoption barrier. Settlement finality in an agentic context is as much a legal and contractual problem as an engineering one. The liability chain in a human-initiated card transaction is well understood: issuer, acquirer, merchant, scheme rules, and consumer protection regulation all play defined roles. In an agent-initiated transaction, the principal-agent relationship introduces a layer that existing frameworks do not cleanly address. Did the agent act within its mandate? Was the mandate itself properly constructed? Was the agent identity authenticated at the point of transaction in a way that satisfies the relevant standard of care?
UK fintechs building in this space should be engaging legal counsel with payments regulation expertise now — not at the point of product launch. The FCA has signalled willingness to engage with novel payment structures through its innovation pathways, and that window of proactive regulatory dialogue is genuinely valuable. Organisations that arrive with a well-considered liability model, a clear audit trail architecture, and a coherent approach to mandate verification will be far better positioned than those treating compliance as a post-build exercise.
The competitive dynamic here is straightforward. The fintechs and infrastructure providers that solve settlement finality for agentic transactions will define the rails on which a significant portion of B2B commerce runs within the next several years. Those that wait for Visa or PayPal to establish global standards will find themselves building on someone else's terms. The UK has the regulatory environment, the technical talent, and the existing payment infrastructure modernisation programmes to take a leading position — but the window for shaping rather than adopting the emerging standard is not indefinite.
For senior decision-makers, the immediate practical step is internal: audit your current procurement and payment workflows to identify where agent automation would deliver the most value, and map the specific settlement and liability questions those workflows would surface. That analysis will tell you precisely which infrastructure gaps you need either to build around or to help solve. The organisations asking those questions now will be the ones with working answers when the market demands them.
What is settlement finality and why does it matter specifically for AI agents?
Settlement finality is the legal and operational point at which a payment becomes irrevocable — neither party can unwind it. It matters for AI agents because existing finality frameworks assume a human initiated and confirmed the transaction, which underpins dispute rights, fraud review windows, and liability allocation. Remove the human and those assumptions break down, creating unresolved questions about who bears loss when an agent transaction goes wrong.
How does Visa's Intelligent Commerce initiative differ from standard API-based payment integration?
Visa's Intelligent Commerce framework is designed to allow AI agents to hold payment credentials and act on delegated authority from a human principal, rather than simply passing payment details through an API at a human's request. The distinction is that the agent itself is treated as an authorised counterparty, not merely a conduit — which requires new identity, credentialling, and mandate-verification infrastructure that standard payment APIs do not currently provide.
What does 'delegated authority' mean in the context of agentic payments, and how is it enforced?
Delegated authority means a human principal — say, a CFO — grants an AI agent permission to transact within defined parameters, such as spending limits, approved vendor lists, or contract value thresholds. Enforcement today relies largely on application-level policy rules rather than verified payment-rail controls, which means a misconfigured or compromised agent can exceed its mandate before any infrastructure-level check catches it. More robust enforcement requires mandate parameters to be cryptographically bound to agent credentials at the payment rail level.
Are there existing UK regulatory frameworks that cover AI-initiated payments?
No current UK regulatory framework explicitly addresses AI-initiated payments as a distinct category. Existing payment regulations — including the Payment Services Regulations 2017 and FCA-supervised conduct rules — apply to the authorised firm processing the transaction, not to the nature of the initiating party. The FCA's innovation pathways offer the most practical route for fintechs seeking regulatory clarity on novel structures, but formal guidance specific to agentic transactions has not yet been issued.
What role could the UK's New Payments Architecture play in enabling agentic transactions?
The New Payments Architecture is designed to modernise the UK's interbank payment infrastructure with more flexible, data-rich messaging standards. In principle, its layered design could accommodate non-human counterparty identifiers and programmable payment conditions — but only if the policy intent and technical specifications are deliberately extended to cover agentic use cases. That requires active engagement from fintechs and industry bodies during the design phase, not retrospective adaptation.
How do smart contracts relate to settlement finality for agent-to-agent commerce?
Smart contracts can encode the conditions under which a payment automatically achieves finality — for instance, releasing funds only when delivery is confirmed, or reversing a transaction if a predefined condition is not met within a set window. Applied to regulated payment rails rather than purely blockchain environments, similar programmable logic could give agentic transactions deterministic, auditable settlement outcomes without relying on after-the-fact dispute processes. Several UK fintechs are actively exploring this approach within the open banking and NPA context.
What audit trail requirements should enterprises expect regulators to demand for agentic payment transactions?
Enterprises should plan for audit trails that capture the agent's identity and credential state at the point of transaction, the mandate parameters it was operating under, the decision logic that triggered the payment, and a cryptographically verifiable chain linking those elements to the final settlement record. Regulators are likely to draw on existing standards for electronic consent and payment authorisation as a baseline, extending them to cover the principal-agent relationship and mandate verification. Organisations that design audit infrastructure proactively will be significantly better placed than those retrofitting it.
What is the practical difference between an AI agent making a payment error and a human making one, from a liability standpoint?
When a human makes a payment error, liability frameworks — consumer protection law, card scheme chargeback rules, and bank error rectification duties — provide relatively clear remedies. When an agent makes an error, the question of whether it acted within its mandate, whether the mandate was properly constructed, and whether the principal exercised adequate oversight all become live legal issues without settled answers. In practice, this means enterprises currently face uncertain liability exposure, which is one of the primary reasons large-scale autonomous agent procurement remains limited to low-risk, low-value transaction types.
How should UK enterprises begin assessing their readiness for agentic payment workflows?
The most practical starting point is a workflow audit focused on high-frequency, lower-value procurement categories where human confirmation adds friction without proportionate control value. For each candidate workflow, map the specific points where settlement finality, dispute resolution, and liability assignment would be triggered if an agent were operating autonomously. That exercise will surface the precise infrastructure and legal gaps your organisation needs to address before deployment, rather than discovering them at the point of a contested transaction.
Will established card networks like Visa and Mastercard ultimately define the standards for agentic payments, or is there space for alternative rails?
Card networks have significant first-mover advantage and the commercial relationships to drive adoption, but their frameworks are being designed to extend existing card infrastructure rather than rethink it. Alternative rails — including open banking payment initiation, account-to-account settlement, and potentially programmable money structures — may offer more architecturally coherent solutions for high-frequency agent transactions where card interchange economics and chargeback structures are poorly suited. UK fintechs have a credible opportunity to develop competitive alternatives, particularly for B2B use cases where consumer protection rules are less constraining.
Get in touch today
Book a call at a time to suit you, or fill out our enquiry form or get in touch using the contact details below