Agentic commerce is happening now. Here's what we've learned.
We’ve been collaborating with others to explore when and how agentic commerce will work. Robin Gandhi is the CPO of Lithic, a leading card issuer that’s already seeing agents use its cards to make purchases. Below, he shares his thoughts on what’s changed, and what needs to change, for agentic commerce to become mainstream.
Last year, I wrote about the opportunity for agentic payments to revolutionize travel bookings, ad spend management, procurement, and more. Since then, we've moved from theory to practice, and we’re working directly with the innovators building AI agents that spend real money using Lithic-issued cards. Here's what we've learned, and where we think this is all going.
Existing trust frameworks aren’t necessarily broken
Putting a payment mechanism into the “hands” of an AI agent raises a question that cuts to the heart of the payments ecosystem. How do we (merchants, cardholders, and issuers) establish trust when there's no human in the loop at the moment of transaction?
The answer could start with the infrastructure we've already built.
Agentic payments are a new execution layer, but they're not really different from the fundamental principles of online commerce. The same foundations that allow people to buy stuff online can apply here if we change the framing a bit.
Know Your Customer (KYC) and Business (KYB): The agent should act on behalf of a person or business, not instead of one. It’s a critical distinction that lets us extend existing identity frameworks into agentic contexts. Agentic workflows are a new way for Lithic customers to access their accounts. By using Persona for KYC and KYB during onboarding, we ensure there's a verified and legitimate entity behind every account and every agent.
Fraud rules and authorization controls: Whether someone gives a card to their agent or their human assistant, rules and controls are essential for building a trustworthy environment. Lithic's Authorization Intelligence controls let people set limits on where, when, and how a card can be used. These programmable controls can reduce fraud risk and create important audit trails.
Disputes and chargebacks: Disputes aren't going away in an agentic world. If I had to guess, I'd expect the volume to grow. Our API-first approach, along with agentic tools for dispute resolution, helps our users manage chargebacks programmatically.
But the frameworks need an update
In addition to reframing the fundamentals, the payments ecosystem needs to reexamine existing assumptions and rebuild certain systems to account for agentic payments:
Net-new data signals: For a long time, fraud models (and the people behind them) were trained to treat bot and headless-server activity as uniformly bad. This will need to change. We’ll need to incorporate new signals for: agent provenance; cardholder intent/mandate; links between agents and accounts, devices, and networks; and whether the human behind the agent is real and well-intentioned.
Step-up authentication: Today, protocols like 3D Secure (3DS) provide a valuable friction point, allowing us to challenge buyers and add assurance during sensitive moments. These authentication protocols need to be updated to enable agents to participate, and we’re already seeing Visa and Mastercard building in this area.
Peering into the future of agentic commerce
Over the past year, we built demos, stood up agentic payment flows, issued highly restricted cards, and then ran into a consistent wall. The cards and accounts worked, but the overall ecosystem wasn’t ready. Merchants blocked the agents because the transaction lacked the device fingerprinting signals that acquiring banks and fraud systems expected.
But that changed recently. With the hyper-adoption of browser-based automation tools like OpenClaw, agents began generating the device signals needed to pass merchant-acquiring fraud checks. Transactions that had been previously blocked started to go through. This doesn’t solve everything related to liability, chargebacks, or fraud, but it does change the game materially.
Example of an agentic commerce transaction
As agentic commerce matures, we expect to see:
Agent identity as a real concept: Right now, agents operate under the identity of the human account holder. Over time, we expect the industry to develop credentialing and attestation standards for agents themselves. Identity verification platforms, like Persona, will be key to this transition.
Programmable trust policies: Cardholders will want nuanced control over their agents. For example, they might want to allow the agent to book flights but not hotels, or limit purchases to a list of approved merchants. This is a product surface we're actively building toward.
New authentication flows built for agents: The 3DS-for-agents problem will likely get solved through a combination of cryptographic agent attestation and risk-based decisioning that assesses agent behavior rather than human responses.
Agentic commerce is in its infancy, but it’s happening. There are real cards, real transactions, and real stakes. And I think it’s only going to become a bigger piece of card-not-present transaction volume over time. The teams building AI agents need payment infrastructure that can keep up: programmable controls, new data signals, step-up flows, API-first operations, and a trust framework built for the way agents actually operate.
If you're building in this space, let’s talk. Reach out to me directly ([email protected]) to learn how Lithic can power your agentic payment workflows or sign up and start building today. And reach out to Persona, which will be a valuable source of trust signals, to explore how their digital identity platform can help enable your agentic AI use case.