Abdul Abdirahman is a principal at F-Prime, a global venture capital firm focused on healthcare and technology.

AI agents are already playing important roles as recommendation and information gathering engines for e-commerce. However, this application only scratches the surface of AI’s potential in online marketplaces. Very soon, AI agents won’t just find a desired product or service for users — they will also autonomously execute the payment. Empowering AI agents to complete transactions without human intervention will unlock a new era of efficiency and convenience in both consumer and business transactions, and represents a new wave of innovation in global commerce. 

The evolution of online payments has mirrored the growth of the internet itself. Early iterations like PayPal operated within closed ecosystems and, as mobile technology advanced, a second wave of mobile payment systems emerged. This era saw a mix of open and closed systems designed to facilitate simpler payment experiences for consumers. Companies like Adyen and Stripe also emerged to create developer-friendly platforms empowering businesses to integrate seamless checkout into their websites and applications. Meanwhile, Asian markets drove the rise of chat-enabled closed-loop ecosystems like WeChat Pay, which blurred the lines between messaging and commerce. These advances — and the ecosystem of tools surrounding these platforms — were all designed to maximize sales conversion in human-orchestrated payments. 

The current state of agentic payments

This is a very nascent field, and most agentic payments startups are focused on B2B and developer use cases. On the consumer front, most startups are e-commerce shopping engines and not attempting to tackle the actual payment (see Operator from Open AI, Perplexity Shop, DayDeam, Deft, and Remark). 

We can imagine a world where these shopping engines are incorporated into voice assistants as purchasing agents. In this scenario, Alexa or Siri would find and purchase goods and services on behalf of users. It remains unclear whether existing players will build their own infrastructure for agentic payments — think Siri + Agentic ApplePay — or if Apple, Google, Amazon and Alibaba will utilize startup-built infrastructure to integrate agentic payments. 

However we get there, we cannot understate the disruptive potential agentic payments hold for e-commerce. Widespread adoption would revolutionize the e-commerce landscape and alter revenue and profit pools across the broader payments space. 

In 2024, consumer and business e-commerce transactions totaled $300 billion and $2.1 trillion, respectively. The global e-commerce market is worth around $6 trillion, the US market is growing at around 8%, and even faster abroad due to increasing digitization. Per the F-Prime Fintech Index, we note that take rates by payment facilitators range from 15 bps to 562 bps depending on size of partner, risk, and geography served. Globally, this means approximately $250 billion in payments could be disrupted by agentic payments.

Challenges to solve

While the opportunity is huge, the path to scalable agentic payments is riddled with structural obstacles. Any solution that can go beyond mere recommendations and actually execute transactions would have to solve the following:

• Financial infrastructure: Our current financial system is designed for human interactions, presenting a critical mismatch for AI agents. So how does an AI agent gain access to traditional banking, payments, and card rails?

• Lack of regulatory clarity: Current rules, regulations, and user onboarding flows would face significant scalability issues if millions of AI agents leverage the existing rails. Furthermore, payment card industry (PCI) compliance regulations designed to protect cardholder data and funds pose a challenge for agentic payments. Developers are restricted from providing their card details to agents due to chargeback guarantees and uncertainty around unauthorized agentic transactions. For agentic payments to flourish within the current ecosystem, more clarity and finetuning of chargeback rules are needed to provide guidance and guardrails.

• Know Your Agent (KYA): There is no robust and reliable system for verifying the identity of AI agents or distinguishing them from malicious bots. There has been movement on this front, however, with Okta’s release of Auth0 AI to help authenticate AI agents and assist with asynchronous user approvals.

• Lack of tailored fraud tools: Current tooling focuses on identifying malicious bots, and relies on recognizing human digital footprints: spending patterns, device/ISP addresses, and how fast we type in card details. These solutions fall short in addressing the unique characteristics of agentic transactions and will inadvertently impede legitimate transactions.

• Tech stack and compatibility issues: Current payment systems are ill-equipped to handle the unique workflows and negotiation interfaces needed by AI agents. Agentic transactions might require recurring payments, microtransactions, or real-time contextual buying decisions, capabilities not efficiently supported by our existing payment infrastructure.

• API usage: Today’s APIs are designed for human developers to engage with and thus not optimized for the agentic world. Many modifications are needed to enable agents to execute complex tasks efficiently. For starters, AI agents require non-deterministic APIs with the capacity to make numerous calls for a single request. This is not a functionality adequately addressed by existing APIs, nor can existing APIs handle the impending 10-100 times  increase in API utilization by AI agents. APIs also need to evolve from static endpoints to intelligent entities capable of managing tasks dynamically.

The above amounts to an inherent incompatibility between established payment systems and the emerging needs of AI agents, necessitating a paradigm shift in financial infrastructure to unlock the full potential of agentic commerce. 

How startups are navigating these challenges

When addressing the limitations of the current payment systems and building foundations for the agentic future to come, startups are focusing on two basic areas: pay-ins and payouts.

Relatively speaking, the process of enabling AI agents to accept payments is technically simple. Protegee AI, for example, offers a PCI-compliant payments API specifically for AI voice agents, allowing them to securely receive payments over the phone. Inbound phone sales, debt collection agencies, and clinics currently use this solution. The relative ease of implementing payment acceptance suggests that new, vertical-focused startups will likely emerge alongside established companies that offer bespoke services tuned to the needs of their target industry. 

However, chargebacks and KYA concerns still remain. 

Payouts, on the other hand, present a more intricate challenge given the regulatory hurdles and security considerations involved in granting AI agents the ability to do mass disbursements. The infrastructure to conduct payouts simply does not exist right now. Startups like Skyfire want to build a Visa-like network for agentic payments; Nevermined is building “PayPal for AI agents”; and Payman AI is enabling AI agents to pay humans for their work — connecting agents with bank accounts to enable fiat-based payments in certain geographies. The latter represents a departure from the reliance on stablecoin rails to facilitate money movement. 

In response to this new use for crypto networks, Coinbase and other scaled Web3 fintechs have developed a platform for agent-to-agent payments and made it easier for developers to spin up AI agent crypto wallets. Meanwhile, Stripe launched an agent toolkit in mid-November along with a framework that grants AI agents access to its API, which will allow AI agents to manage funds, facilitate support operations and implement usage-based billing. Some agentic payments startups are leveraging Stripe for payment processing service and to handle funds and accounts. 

The immediate future

We are very excited about the potential for agentic payments, but clear-eyed about the massive challenges that must be overcome to build that future. In 2025, we are closely watching startups building key infrastructure such as KYA and agentic fraud tooling; AI payment orchestration; AI stablecoin wallets; and specialized APIs like Glide AI and BlindPay. These tools will be crucial to create seamless and secure agentic transactions over the coming years.