Beyond the Checkout Button: The Age of Agentic Payments

Beginning of the Agentic Commerce

Sep 12, 2025

It's clear the future of e-commerce isn't about human-to-machine transactions anymore. It's about AI-to-AI. For those of us building the products that power this world, this is a pivotal moment that requires a complete paradigm shift in how we think about payments.

The first step to understanding agentic payments is to distinguish them from their older, less intelligent cousins: Machine-to-Machine (M2M) payments.

Think of M2M like a pre-programmed robot. You set a rule, say, "If my smart home needs new coffee beans, order them from Amazon." The coffee maker’s sensor hits a pre-set threshold, it sends a signal, and the payment is triggered. It’s an efficient, rules-based system, but it’s essentially just following a script.

Agentic payments, on the other hand, are the evolution of this concept. An AI agent is not just reacting to a pre-programmed rule—it’s reasoning, learning, and acting with a purpose.¹ It's a "Delegated Commerce" system.

Instead of your coffee maker simply re-ordering the same beans, a personal AI agent might:

Identify that you’re low on beans.
Search for the best-rated, ethically sourced coffee from local roasters.
Negotiate a discount based on a flash sale.
Check your calendar to see if you’ll even be in town to drink them.
Order the coffee and pay for it autonomously, all without you lifting a finger.

This is the key difference: autonomy and goal-setting.³ As industry thought leader David Birch puts it, we're moving from a world of M2M to one of Robot-to-Robot (R2R) payments, where the "smart wallet" becomes the central, orchestrating intelligence. For a payments PM, this means your customer isn't just a single person anymore, but an intelligent, dynamic AI agent actor that needs new tools, new rails, and new safeguards.

The Challenge and the Opportunity

This is a moment of unprecedented change. So, as you head into your next product roadmap meeting, don’t just think about how to build a better checkout flow. Start thinking about how to build a better wallet for a world where payments are delegated, and commerce is truly autonomous.

The Payments Paradox: When Bots Become Your Best Customers

For decades, the foundation of digital commerce security was a simple heuristic: "bot-like" behavior is bad.⁴ A flurry of rapid, high-frequency transactions was a clear red flag for fraud.¹ But with agentic commerce, that's no longer the case. A legitimate AI agent operates at machine speed and scale, making the very behavior that used to signal criminal activity now a feature of a trusted user.⁴

This requires a complete overhaul of our fraud prevention strategies. The old "Know Your Customer" (KYC) model is becoming obsolete, giving way to a new paradigm: "Know Your Agent" (KYA).⁴ Fraud experts are already reporting being targeted by AI-enabled fraud, with one study showing 65% of experts have seen such attacks.⁴ This new frontier of fraud is characterized by sophisticated attack vectors that go beyond traditional phishing or malware. These include data poisoning—where an attacker subtly manipulates an agent's training data to influence its decisions—and supply chain attacks, where malicious code is injected into third-party libraries an agent uses.⁶

The challenge for product managers is to build a new trust infrastructure that can distinguish a "good" AI bot from a "bad" one.⁴ This moves the security primitive from human-provided credentials like passwords and biometrics to a new system of cryptographic delegation. Think of a chain of trust: a User's ID Token proves the human's identity, an Agent ID Token acts as the agent's digital passport, and a Delegation Token cryptographically links the two, explicitly defining the scope of the agent's authority.⁷ Your job is to create the primitives that ensure this chain of trust is unbreakable.

This shift also introduces new problems for dispute resolution. If a consumer’s agent makes an erroneous purchase, does the traditional card dispute process, which assumes a human is responsible for every purchase, still apply? ⁸ The industry must now solve for a "liability gray area" where an agent’s transaction is technically authorized but ultimately defies user intent, and the causal link between the agent's actions and a chargeback is unclear.¹⁰

The Regulatory Headaches

The autonomous nature of AI agents and their ability to act without direct human intervention introduces a significant, unresolved challenge of liability, creating a "trust vacuum" that inhibits adoption at scale.¹⁰ The core of the problem lies in the fact that AI agents do not possess intention (

mens rea) in the way that humans do.¹³ This absence of a key legal concept breaks traditional frameworks like tort and product liability laws, which are often predicated on the intent of the actor.¹³ When an agent makes an erroneous purchase that technically follows authorization but ultimately defies the user's intent, it creates a "liability gray area."¹⁰

Legal scholars and emerging case law suggest that liability will likely fall on the human who deployed the agent, the developer, or the manufacturer of the system.¹³ This is compounded by the "agentic loyalty problem," a conflict where an AI agent could be loyal to the platform that deployed it rather than the user's best interests.¹⁰ For example, an agent could choose a more expensive item from a partner company even when a cheaper, more suitable option exists elsewhere.¹⁰ This direct conflict of interest violates a core legal duty in agency law and highlights how the incentives of the e-commerce industry are structurally opposed to a user's best interest, creating a fundamental breach of trust at the heart of the agent-user relationship.¹⁰

The Emerging Agent Economy: From Credit Cards to Micropayments

The rise of agentic commerce also exposes a fundamental incompatibility between the new machine economy and our legacy financial infrastructure. Our current payment systems—credit cards, subscriptions, and batch settlements—are built for human-centric, high-value, low-frequency transactions.¹⁶ They are ill-suited for a world where autonomous agents need to pay for individual API calls, data, or computing resources on a "pay-per-use" basis, often for fractions of a cent.¹⁶

The economic friction of a standard credit card transaction is simply too high for this type of high-frequency, low-value interaction.¹⁶ This has created a direct demand for new, purpose-built payment rails that can handle secure, fast, and cost-effective machine-to-machine transactions.¹⁶

This is where a new wave of crypto-native technology comes in. Blockchain-based micropayment systems are emerging as the foundational infrastructure for this new economy.¹⁸ For example, Coinbase has launched x402 Bazaar for USDC micropayments, which offers a settlement path in just 200 milliseconds with nominal gas fees.¹ This low-latency, high-granularity system is exactly what is needed for agents to purchase data or rent computing resources without human intervention.¹⁶ Smart contracts—self-executing agreements on the blockchain—will form the backbone of this new value exchange, automatically triggering payments when predefined conditions are met.²⁰

Beyond payments, this new infrastructure provides core primitives that traditional systems lack: a secure, immutable ledger, verifiable identities, and smart contracts for autonomous value transfer.¹⁸ In a decentralized, multi-agent world, an agent needs a way to prove its identity and verify that another agent is trustworthy.¹⁸ Concepts like "Agent-Based Tokens" (ABTs) are being developed to bundle an agent's identity, reputation, and financial security into a single, on-chain primitive, enabling a self-governing and self-monetizing digital workforce.¹⁹

Final Takeaway

The next era of e-commerce isn’t about building a slightly better version of what we have today. It's about designing an entirely new system for an entirely new kind of user. As payment product managers, you are at the forefront of this revolution. The challenges are significant—from re-engineering fraud detection for bots to building new payment rails for a machine economy—but the opportunity is immense. The future belongs to those who are bold enough to build a better wallet for a world where commerce is truly autonomous.

Sherman’s Substack

Discussion about this post

Ready for more?