Agentic Payments-Part 2
How Visa, Mastercard, PayPal, and Stripe Are Each Building the Future of Autonomous Payments
In my previous article, I outlined the market opportunity behind agentic payments. This time, I’m diving into how major incumbents and network providers are evolving their strategies and infrastructure to enable this shift. The focus will be on Visa, Mastercard, Paypal, Stripe and the different approaches they are taking to facilitate this new wave in the payments world.
Visa:
Visa’s focus is on core infrastructure and to make the existing network rails more agent ready and friendly.
What was announced or launched?
On April 30, 2025, Visa unveiled Intelligent Commerce, designed to support agent-initiated commerce. It enables AI agents to securely hold and use Visa credentials through agent-specific tokens and passkey authentication.
What does it do, and what’s the underlying technology?
Agent-specific network tokens replace traditional 16-digit PANs with unique, revocable identifiers assigned directly to AI agents. These tokens are protected by passkey-based authentication and linked to Visa's fraud detection infrastructure, allowing agents to browse, compare, and complete transactions without exposing sensitive user data.
What is passkey authentication?
Passkey authentication is a secure login that replaces passwords with cryptographic credentials stored on a device, such as a smartphone or computer. In Visa’s case, this ensures that only a verified agent using a device tied to the passkey can access and use the tokenized credentials to initiate a payment.
Where does this fit in with the broader payments/fintech ecosystem?
Sits at the infrastructure and identity layer, Visa is modernizing the payment “rails” to accommodate agentic activity securely and at scale.
Visa is playing defense around its core network while simultaneously making a land grab to become the default payment layer for consumer-facing agents.
What’s their competitive advantage?
Their existing relationships with issuers and acquirers create a powerful moat by embedding Visa’s infrastructure directly into banks and merchant systems making displacement by competitors technically complex, commercially risky, and operationally expensive.
How is it differentiated?
Its focus is on ensuring universal, interoperable agentic tokens and driving cross-industry trust.
Unlike Mastercard, which emphasizes auditability and programmable controls for regulated use cases, Visa focuses on enabling trusted, scalable agentic transactions across the entire network prioritizing acceptance and interoperability.
What’s the strategic intent?
This is a foundational, defensive, and long-game move to remain the default transaction layer in the age of AI, ensuring Visa credentials remain usable even when no human is involved at checkout.
Who’s the target market or end-user?
AI agents built by developers at OpenAI, Microsoft, and Samsung;
Financial institutions issuing cards by integrating Visa’s agent token architecture into their authorization layers, enabling customer accounts to delegate transactions to verified AI agents with tokenized, bank-approved credentials
Large merchants integrating agent-based shopping.
Mastercard:
Mastercard’s focus is on providing risk and safety solutions for enterprises and also tries to act as control layer
What was announced or launched?
On April 29, 2025, Mastercard launched Agent Pay, a flagship solution within its broader Agentic Payments Program. The core innovation: Agentic Tokens are programmable, payment credentials designed for AI agents that embed usage rules, permissions, and traceability into every transaction.
What does it do, and what’s the underlying technology?
These tokens integrate directly into AI agents and provide full audit trails, making every transaction visible and governed through enterprise-grade permissioning and traceability. These tokens integrate with AI agents embedded in chat apps or B2B workflows, enabling secure, pre-authorized payments with a full audit trail.
Where does this fit in with the broader payments/fintech ecosystem?
It sits at the enterprise orchestration and risk layer, enabling institutions to authorize payments not just by identity but by programmable logic. Unlike traditional rails that process static payment credentials, Mastercard’s model enforces dynamic controls, making it ideal for high-risk, high-value agentic transactions.
What’s the USP?
Mastercard is creating a compliance-first platform for agentic commerce, wrapping every agent transaction in enterprise-grade guardrails. By enforcing rules at the credential level, it ensures agents act within traceable, auditable limits, making it the go-to infrastructure for B2B, government, and regulated sectors. Mastercard is building programmable, permissioned logic on top of the rails.
What’s the strategic intent?
Positioning itself as the control layer. This is both a wedge into enterprise AI and a hedge against regulation by proactively embedding governance.
Unlike others building infrastructure or developer tools, Mastercard offers the strongest compliance posture, embedding enterprise-grade tooling directly into payment flows.
Who’s the target market or end-user?
Large enterprises, B2B SaaS platforms, regulated sectors (healthcare, logistics, government), and financial institutions.
PayPal:
PayPal is trying to be the SMB enablement platform
What was announced or launched?
At Dev Days 2025 and in Q1 earnings, PayPal unveiled its Agent Toolkit and expanded Model Context Protocol (MCP) Servers. These tools are designed for quick, low-code implementation of agent-driven payment experiences and to focus on rapid developer adoption and SMB enablement.
What does it do, and what’s the underlying technology?
The Agent Toolkit is a developer library that wraps PayPal’s suite of products like Checkout, Payouts, Subscriptions, Invoicing, Shipping, Disputes into a natural language-accessible API. MCP enables LLMs to securely call PayPal’s systems without complex setup.
Where does this fit in with the broader payments/fintech ecosystem?
This sits in the application and merchant enablement layer, especially for SaaS tools, e-commerce apps, and AI assistants that need embedded payments but lack in-house payment infra.
PayPal is targeting the long tail of SMBs and creators, aiming to grab market share before larger networks move downstream. Unlike others building infrastructure or governance models, PayPal offers the quickest path to live payments for lightweight use cases and non-technical users.
What’s the USP?
PayPal’s plug-and-play speed, large wallet base (400M+ users), and developer-friendly onboarding make it the fastest way for small to mid-sized companies to experiment with agentic payments. Their focus is on accessibility and ease, removing technical barriers to enable agentic commerce with minimal lift for small businesses and developers.
How is it differentiated?
Where Visa and Mastercard prioritize rails, PayPal focuses on user experience and adoption velocity, especially for SMBs and non-technical builders.
This is a wedge product to reclaim developer attention and SMB relevance in a space increasingly dominated by Stripe and newer infra players.
Who’s the target market or end-user?
SaaS providers, workflow automation tools, solo developers, small to mid-sized businesses, and e-commerce platforms looking for ready-to-integrate agentic payment flows.
Stripe:
Stripe focusses on being the developer-first platform for agentic commerce.
What was announced or launched?
Between May–June 2025, at Sessions 2025, Stripe showcased its Order Intents API, Commerce Agent API, and enhanced Radar ML fraud tools. Also announced: stablecoin wallets, virtual cards, global payouts, and chat-native payment features.
all focused on developer experience and programmable automation.
What does it do, and what’s the underlying technology?
Stripe’s Order Intents API lets AI agents handle checkout steps like reserving inventory, calculating tax, and confirming payment in one streamlined call. It also includes upgraded fraud detection (Radar ML), supports stablecoin wallets, and enables global payouts all built for automation and speed.
Where does this fit in with the broader payments/fintech ecosystem?
It fits within the programmable middleware and developer tooling layer, providing flexible APIs that sit between agent applications and underlying card/payment networks. Stripe enables builders to orchestrate agentic commerce without relying on fixed UI or rigid enterprise controls. Its approach focuses on rapid developer enablement, automation tooling, and modularity, targeting fast-moving teams who want to prototype, iterate, and scale custom payment flows without friction.
What’s the USP?
Stripe treats payments as software infrastructure. Its tooling is modular, code-native, and Git-controlled, giving developers end-to-end flexibility and speed. Everything is API-first and Git-native.
How is it differentiated?
Stripe’s strength is its deep developer love and a clear focus on automation, iteration speed, and flexible architecture for modern dev teams.
What’s the strategic intent?
Stripe is making a land grab for agentic app developers and positioning itself as the default API layer for AI-native commerce stacks.
Who’s the target market or end-user?
Startups, AI-first product teams, fintechs, SaaS platforms, and any developer building custom payment flows with AI agents.
Now that we’ve seen how each player is positioning itself, it’s worth stepping back to assess how much of this is real, what’s still missing, and who’s best placed to lead as agentic payments move from theory to practice.
This is early-stage positioning, not operational reality. All four companies are essentially announcing developer tools and APIs, the actual AI agents that would use these systems barely exist at scale.
What's Actually Working: The solid technical foundations are being laid. Tokenization, fraud detection, and programmable controls are proven technologies being repackaged for AI use cases.
Major Gaps:
Agent Ecosystem: Few production AI agents actually need autonomous payment capabilities today
Merchant Adoption: Most retailers aren't equipped to handle agent-initiated transactions
Regulatory Clarity: No clear framework for AI agent liability and compliance
User Trust: Consumers aren't ready to delegate spending decisions to AI
Adoption Outlook: Currently low, but accelerating. We're in the "if you build it, they will come" phase. Success depends on AI agents becoming genuinely useful for commerce tasks like price comparison, subscription management, and repeat purchases.
Why Visa might have an early advantage?
It’s still early, but Visa is shaping up to be the early front-runner. Not because of the speed but the approach to make current rails more agentic focused.
Default status: Banks already issue Visa cards; extending this to AI agents is a natural evolution, not a platform switch. Financial institutions can enable agent payments through existing card programs without overhauling their core systems or renegotiating partnerships.
Zero merchant friction: Unlike competitors requiring new APIs or integrations, Visa's approach works with existing payment terminals and e-commerce checkouts. Any merchant that accepts Visa cards today can immediately accept agent-initiated payments—no technical upgrades, no new contracts, no implementation costs.
Regulatory Head Start: Existing compliance frameworks and bank relationships reduce regulatory risk compared to newer payment models. Visa operates within established banking regulations, anti-money laundering requirements, and consumer protection laws that regulators already understand and trust.
Visa's strategy is less about innovation and more about inevitability, they're making agent payments work within the existing financial ecosystem rather than forcing merchants, banks, and regulators to adopt entirely new systems. This approach leverages decades of built infrastructure and regulatory relationships, creating a path of least resistance that competitors can't easily replicate.
To conclude, agentic payments are in the early stages, and much of what exists today is groundwork. Long-term success will depend on how well these solutions integrate into real-world systems and workflows.
Wow. This is so deeply researched and insightful!