Agentic AI and the programmable future of digital payments

Ignacio E. Carballo, Head of Alternative Finance and Senior Consultant at Payments and Commerce Markets Intelligence (PCMI), analyses how Agentic AI is impacting digital payments.

From one-click to autonomous payments

For years, digital payments have been about reducing friction, from swiping a card to one-click checkouts. Yet even the slickest ‘Pay Now’ experiences have required a human at the helm to initiate each transaction. That is beginning to change. 

A new paradigm is emerging in which Agentic AI (autonomous digital agents) can transact on behalf of users and businesses. In this model, money shifts from a passive instrument to an active participant in commerce. Consumers might soon send a bot to do the buying, ‘while going to do more interesting things with their life’, as one Mastercard Chief Services Officer, Craig Vosburg, quipped. 

In other words, instead of just showing recommendations, AI agents can now search, decide, and pay. This convergence of AI and payments is creating a financial architecture that is not only faster and more global but deeply programmable, where transactions carry embedded logic and can execute autonomously (as I explained in a previous article, ‘The programmable future: how stablecoins are rewriting global finance in the AI era’). 

Early examples of this shift are already in the market. Visa’s CFO, Chris Suh, said at a William Blair investor conference that generative AI (GenAI) could transform commerce as profoundly as ecommerce did a generation ago. Online platforms are experimenting with ‘no-click’ purchases: Amazon is testing a ‘Buy for Me’ AI feature to let its app automatically order items for users, and OpenAI, along with Stripe, launched an ‘Instant Checkout’, so ChatGPT can not only find a product but also buy it on behalf of a user. 

Meanwhile, fintechs and payment networks are racing to enable these agent-driven transactions at scale. All this signals that digital payments are evolving from a user-driven process to an agent-driven one, a fundamental leap akin to moving from manually driving a car to relying on a self-driving vehicle. 

When hype meets execution: AI agents enter real payment flows

In 2025, a flurry of industry announcements revealed just how seriously incumbent players are taking agentic payments. 

Mastercard rolled out its new ‘Agent Pay’ platform, designed to enable trusted AI agents to register and securely make payments on behalf of users. The idea is simple but powerful: an AI shopping assistant could have its own ‘verified ID’, so that merchants and issuers know a legitimate agent is initiating a purchase, not a fraudster’s bot. To build confidence, consumers retain complete control over what an agent can spend on their behalf, much like setting a credit limit or parental controls, but for your AI helper. 

Not to be outdone, Visa unveiled a similar effort, dubbed a ‘Trusted Agent’ protocol, and teamed up with partners like Stripe to embed agentic commerce into the payments stack. Visa envisions ‘millions of people...transact[ing] in an AI-driven world in a way that is secure and trusted’ under these new frameworks. 

Crucially, payment firms are not working in isolation; they’re collaborating across the ecosystem. Mastercard’s programme, for instance, relies on its tokenisation expertise (the same tech behind card-on-file and recurring payments) and involves partners from bigtech to fintech – Microsoft for AI infrastructure, Checkout.com and Braintree for merchant integration, and even IBM to explore B2B use cases. 

PayPal is also joining the fray. In April 2025, PayPal introduced its ‘Agentic Commerce Toolkit’ for developers, allowing easy integration of PayPal payments into AI agent workflows. By October, PayPal went a step further, announcing it would integrate Mastercard Agent Pay into the PayPal wallet, connecting hundreds of millions of consumers and millions of merchants into the agentic payments ecosystem. 

The trend is clear: from card networks to fintech platforms, the industry is coalescing around agent-driven commerce. Each of these initiatives brings us closer to a world where AI-driven transactions are routine. 

New ‘agentic’ payment features as of December 2025

Company	Pilot Name	Explanation
	Creator Agent Pilot	Visa’s pilot automates payments and reminders for creators using AI agents.
	Agent Pay/Agentic Pay	Mastercard enables secure, tokenized payments by AI agents for consumers.
	Agentic Toolkit	PayPal’s toolkit allows automated purchases within conversational interfaces.
	Agentic Commerce Integration	Stripe integrates agentic commerce for automated loyalty and payments.
	AP2 (Agent Payments Protocol)	Google’s protocol authorises agent-driven payments with cryptographic mandates.
	Instant Checkout (ChatGPT)	OpenAI lets users buy products directly within ChatGPT conversations.
	Agentic Payments with UPI	Razorpay enables pre-authorised, automated payments via UPI in India.
	Agentforce	Salesforce automates B2B reordering and customer service with AI agents.
	Smart Payment Agent Platform	Galileo offers API-driven agent payments for bill pay and expense management.
	Agentic Commerce Integration	PwC integrates agentic commerce tools for loyalty and payment automation.

Source: PCMI 

This momentum is about real use cases. AI-powered agents promise to boost sales by making shopping effortless (more sales mean more network fees), and to unlock new payment flows which would be impossible in a purely manual world. Consider subscription services that negotiate the best renewal deals, or travel agents that continuously scan for price drops and re-book you, both doing so automatically. Stripe’s CEO recently described these agents as ‘always-on commercial copilots’ that could handle everything from checkout to fraud screening autonomously, a natural extension of digital payments into the realm of digital assistants, while Amazon’s CEO, Andy Jassy, emphasised that GenAI will ‘reinvent virtually every customer experience we know, and enable altogether new ones about which we’ve only fantasized’. 

Beyond bots at checkout: AI across the payments stack

Much of the buzz has centred on consumer-facing shopping bots, but the impact of Agentic AI on payments runs far deeper. In practice, GenAI and machine learning (ML) are permeating every layer of the payments value chain, often in less flashy but equally transformative ways. 

According to a recent BCG report on the impact of GenAI in payments, payments companies initially chased AI for efficiency gains, but are now leveraging it to drive revenue, enhance customer experience, and reduce risk across the enterprise. The entire payment stack is in play. This means AI is boosting productivity in core business functions, from software engineering and customer support to marketing and sales, and revolutionising payment-specific workflows, like fraud prevention, compliance, underwriting, and collections. 

The numbers speak to this broad impact. Some large financial institutions already project billions in value from AI. RBC (Canada’s biggest bank) expects its AI investments to deliver USD 700 million to USD 1 billion in enterprise value by 2027, expanding the capabilities of its top performers to 80% of staff via AI. JPMorgan anticipates USD 2 billion in gains as it equips 200,000 employees with GenAI co-pilots. And those are just the top-line figures. 

Drilling down, AI is attacking inefficiencies throughout operations. For example, GenAI can handle routine customer inquiries and disputes via chat or voice, escalating only the complex cases. BCG found that a well-orchestrated AI co-pilot programme in customer service can contain 75–85% of inquiries entirely in self-service mode, dramatically cutting resolution times and operating costs. Similarly, in fraud detection, AI systems are sifting through transactions to flag anomalies with greater precision, reducing false positives and freeing human analysts to focus on truly suspicious cases. In underwriting and risk management, AI models ingest far more data points than any credit officer could, potentially boosting approval rates while controlling risk. 

In short, agentic and intelligent systems are becoming the adaptive glue across the payments ecosystem, making processes not only faster, but smarter and more predictive. 

This enterprise-wide deployment is redefining how payment firms organise themselves. Many are establishing AI task forces or centres of excellence to coordinate these efforts (no one wants 100 disconnected AI pilots). A key insight from the field is that orchestration matters: the greatest gains come when AI isn’t applied in silos but woven into end-to-end workflows. Think of automating a collections cycle: an AI agent could detect early signs of customer delinquency, initiate outreach via the customer’s preferred channel, negotiate a payment plan, and even schedule follow-up payments, all without human intervention. That requires linking data, decisions, and payment execution in a seamless loop. 

Early adopters are tackling such flows and seeing tangible results. As one case study exemplified, when AI orchestrates the hand-off between self-service bots and live agents in a contact centre, companies achieved double-digit cost reductions and higher customer satisfaction. The takeaway is that agentic AI isn’t only about new customer-facing products – it’s a catalyst for re-engineering the entire machinery behind payments. Institutions that harness it across both front- and back-office are expected to gain a significant competitive edge in efficiency, agility, and insight. 

Snapshot: agentic commerce – from today to tomorrow

As Agentic AI matures, it’s critical to recognise that we’re not leaping into a fully autonomous future overnight. Instead, we are transitioning through clearly observable stages. 

Multiple current systems operate as rule-based assistants: reactive, context-light, and dependent on user inputs. The shift toward true agentic commerce involves progressive layers of autonomy, personalisation, and transactional independence. 

The table below summarises this evolution, contrasting the present landscape with the capabilities rapidly taking shape. 

Source: Adapted from CMSPI – Travel & Payments Agentic Commerce 101 

Yellow lights: managing new risks and challenges

The road to autonomous finance is promising, but not without hazards. As payments become more intelligent and automated, trust and governance take centre stage. 

One immediate challenge is ensuring that an AI agent does exactly what it’s authorised to do and nothing more. Banks and networks are well aware that, if a bot goes rogue (or is hijacked), the fallout could be severe. That’s why the current wave of agentic payment solutions emphasises ‘guardrails’: verified agent identities, spending limits, and user approvals for high-value actions. 

Visa and Mastercard’s protocols, for instance, require that AI agents be registered and cryptographically authenticated as bona fide representatives of a customer. Cloudflare, which is helping secure these flows, notes that merchants need ways to distinguish AI shoppers from malicious bots and confirm that an agent truly represents a legitimate customer. Robust authentication and tokenization at every step will be essential to maintain trust as we hand over more payment authority to machines. 

Fraud is another flashing yellow light. Fraudsters will undoubtedly seek to exploit AI-driven processes, from prompt manipulation (tricking an agent into unauthorised purchases) to synthetic identities that fool automated know your customer (KYC) checks. The industry will need new trust frameworks to verify AI-initiated transactions and to detect when an AI agent itself might be compromised. This may include AI observing AI – meta-AI systems that monitor agent behaviour for anomalies, much like fraud systems monitor human spending patterns. Early trials are already exploring this territory. Mastercard’s Agent Pay framework, for example, is being piloted with PayPal to refine how agents and merchants interact securely, with a focus on data sharing and verification, without violating privacy. 

There’s also a human factor. Will consumers (and businesses) trust an invisible algorithm with their money? Surveys show enthusiasm for automation, but trust is earned slowly. One single mischarge or errant payment by AI could sour a user’s confidence for a long time. Payment providers must therefore invest heavily in user education and control, providing clear dashboards of what one’s AI agents are doing, easy ways to intervene or override, and liability assurances if things go wrong. 

Regulatory clarity will help here. Just as stablecoins faced a regulatory evaluation, AI agents in finance may attract new guidelines on transparency, auditability, and accountability. Regulators will want to know who is liable if an AI agent makes an unauthorised transaction. The user, the provider of the agent, or the payment network? These questions are now surfacing, and forward-thinking institutions are engaging with policymakers to shape sensible rules. 

The bottom line is that navigating these challenges is as much about policy and culture as it is about tech. These flashing yellow lights aren’t reasons to stop but merely signals to proceed carefully, with eyes wide open. 

The Agentic AI impact/readiness matrix – a framework for strategists

To harness Agentic AI effectively, payment leaders should evaluate opportunities not just by impact but also by readiness. Not all processes are equally ripe for AI automation, and not all organisations are equally prepared. One useful approach is to map initiatives on a simple impact vs. readiness matrix: 

• High impact, high readiness: ‘quick wins’– these are use cases that can deliver significant value and for which technology (and your data/governance) is mature. For example, automating Level-1 customer inquiries with AI chatbots might fall here if you already have a robust chatbot platform and ample training data. Prioritise deploying these immediately to capture value and learning.
• High impact, low readiness: ‘strategic bets’ – these opportunities (say, an AI-driven risk scoring engine that could reinvent credit underwriting) promise high payoff; however, they need further groundwork, of the likes of cleaner data, model validation, or regulatory approval. These should feed into your innovation roadmap. Invest in pilots and partnerships (e.g., with fintechs or AI vendors) to raise your readiness over time.
• Low impact, high readiness: ‘optimise or postpone’ – here lie automations that are easy to do but yield modest benefits. For instance, an AI tool to auto-generate internal reports might be straightforward (high readiness) but not game-changing (low impact). Implement these opportunistically, as they can still boost efficiency but shouldn’t distract from higher-impact projects.
• Low impact, low readiness: ‘backlog’ – these are currently neither particularly valuable nor feasible things and deserve minimal resources. Monitor them occasionally, as impact or readiness may change, but focus elsewhere for now. 

Categorising potential AI initiatives in this way enables institutions to sequence their agentic AI journey: attacking high-value, doable projects first, while laying groundwork for more ambitious transformations. The matrix also helps cross-functional teams communicate, aligning tech, business, and compliance stakeholders on why certain projects take precedence. Each organisation’s matrix will look different, but the exercise instils discipline, forcing a frank assessment of ‘Where are we ready?’ vs. ‘Where do we need to prepare?’. 

Another dimension of readiness is cultural and organisational. Companies should assess their AI maturity level:

• Is our data infrastructure up to par?
• Do we have the talent to develop and manage AI systems?
• Are our risk and compliance teams equipped to oversee AI decisions? 

For instance, a bank may rate itself highly ready in fraud analytics (thanks to years of ML use), but less ready in AI-driven customer interactions (due to regulatory uncertainty or a lack of conversational AI expertise). Identifying these gaps is the first step to addressing them – whether through hiring, training, or partnering with specialists. In sum, a strategic framework that gauges both impact and readiness can turn the nebulous promise of agentic AI into a concrete action roadmap, ensuring that as you embrace the programmable future, you do so in a prioritised, pragmatic way. 

Final thoughts: preparing for an agentic payments era

In this fast-emerging landscape, one thing is evident: agentic AI will strongly reshape digital payments, not by replacing the existing ecosystem, but by supercharging it. Just as stablecoins injected programmability into money, AI agents are injecting intelligence into the act of payment. The institutions that thrive in this new era won’t be those that simply observe the trend, but those that embrace and shape it. 

From my perspective, we are at a rare strategic inflexion point, reminiscent of the early days of mobile payments or the Internet itself, when new leaders can emerge by riding a structural shift. Established players have huge advantages (trust, customer base, regulatory know-how), but they must move decisively to avoid being outpaced by more agile upstarts or bigtech entrants. 

So, what can payment service providers, fintechs, and banks do now to position themselves for an agentic future? A few guiding imperatives stand out: 

• Empower AI as a core user: treat AI agents as a new customer category. Just as you design products for mobile or web users, design for AI as a client. This means exposing your services via APIs autonomous agent can consume and thinking through user experiences where the ‘user’ is an algorithm. Companies like Stripe and PayPal are doing exactly this – opening their platforms so AI agents can initiate and complete transactions fluidly.
• Build trust into every layer: security, compliance, and transparency are not add-ons; they are foundational. Develop rigorous verification methods for agents (digital certificates, AI ‘licences’), robust fraud monitoring tuned to AI-driven patterns, and clear user consent mechanisms. Ensure every agent-initiated payment is traceable and auditable, as earning user and merchant trust in agentic payments will be a differentiator.
• Upskill and adapt your talent: your human teams are the ones training, supervising, and collaborating with AI agents. Invest in training employees to work alongside AI – whether it’s customer support reps learning to leverage AI suggestions or risk analysts interpreting AI flags. Encourage a culture where AI is seen as a ‘co-pilot’ rather than a threat. This also means updating your governance: have cross-functional committees to review AI outcomes, biases, and ethical considerations regularly.
• Start small but start now: don’t wait for perfect clarity. Instead, identify a few high-impact, manageable pilot projects (refer to the impact/readiness map) and get the ball rolling. Maybe it’s an AI agent to simplify chargeback handling, or a pilot with a trusted merchant leveraging Visa’s or Mastercard’s agent protocols. Use these experiments to learn – both the technical kinks and the customer reactions. By building proof-points now, you’ll be ready to scale when agentic technology matures further.
• Collaborate to set standards: the future of agentic payments will be co-created. Engage in industry consortia, sandbox initiatives, or partnerships (much like the PayPal-Mastercard collaboration). Shared standards for things like agent authentication, data exchange, or dispute handling will accelerate adoption for everyone. No single player can tackle all challenges alone; however, collectively, the industry can establish the ‘rules of the road’ for this new commerce model. 

The rise of Agentic AI, coupled with programmable digital money, points to a financial system where transactions are not just faster, but smarter and more autonomous than ever. In this system, value moves at machine speed, and financial workflows can adjust in real-time without manual input. 

For institutions willing to act, the opportunities are immense: richer customer experiences, efficiency gains, and new revenue streams. Yet, success will require being bold and proactive today. By laying the groundwork (technologically, organisationally, and strategically), banks and fintechs can ensure they don’t just adapt to the coming changes but lead in the agentic payments era. 

The businesses that seize this moment will help define how money moves and interacts in the decades to come, solidifying their place in the next chapter of digital finance. 

About the author

Ignacio E. Carballo is the Head of Alternative Finance and a Senior Consultant at PCMI and has 15 years of experience as a recognised academic and thought leader in financial inclusion, alternative finance, crypto, and blockchain. Argentina-native, Australia-based, Ignacio is also the Director of the Centre of Alternative Finance for Latin America (Universidad Católica Argentina, Business School), a Magister in Economic History and Economic Policies (Universidad de Buenos Aires), and a Magister in Financial Inclusion and Microfinance (Universidad Autónoma de Madrid).

About PCMI