Lithic says agentic payments could push virtual cards into B2B controls

Agentic payments, Lithic’s Robin Gandhi told PYMNTS, are forcing online commerce to rethink systems built to stop nonhuman behavior at checkout. The shift matters because payments infrastructure has long treated bot-like activity as a fraud warning, while AI agents may increasingly act with authority from a consumer or business.

Gandhi, Lithic’s chief product officer, said much of ecommerce technology was designed to block activity that appeared automated or detached from a known device. In an interview with PYMNTS, he said the industry now has to distinguish between malicious automation and delegated purchasing by legitimate software.

That distinction could alter where payment controls sit. Rather than relying mainly on checks at the point of authorization, businesses may need to define what an agent is allowed to buy before a transaction is attempted, Gandhi said.

What are agentic payments?

Agentic payments are transactions initiated by software agents acting under delegated authority from a person or organization. In practice, the payment system must assess both whether the agent is authenticated and whether the purchase fits the limits set by the party that authorized it.

Gandhi said consumer examples are easy to picture, such as an agent buying shoes, booking travel or finding household goods. He told PYMNTS, however, that the more durable opportunity may be in business-to-business payments, where companies already use rules and policies to govern routine procurement.

Businesses regularly purchase recurring supplies, services and inputs. Gandhi said that makes B2B spending a stronger candidate for limited autonomy than consumer shopping, where buyers may prefer to compare options and make the final selection themselves.

For finance and procurement teams, that changes payment control from a post-purchase approval process into a permission framework set in advance. Gandhi told PYMNTS that autonomous purchasing is likelier to emerge first in B2B products and services where a company is comfortable allowing replenishment when supply falls.

Why Lithic sees virtual cards as a control layer

Gandhi identified virtual cards as a practical instrument for agent-led spending because they can isolate the underlying funding source and impose transaction limits. He told PYMNTS that virtual cards can support agentic commerce because software agents need a controlled way to spend.

Virtual cards can be configured for single use or restricted by amount, merchant, timing and other rules. That structure can limit losses if an AI agent makes an error, is compromised or produces an unreliable instruction, according to Gandhi’s comments to PYMNTS.

The card model also avoids a major adoption problem for merchants. Gandhi said merchants already accept cards, so agent-initiated card payments do not require them to adopt a new settlement asset or rebuild treasury processes.

PYMNTS reported that stablecoin-native payment models may require merchants to accept new assets and decide how to handle them after settlement. Lithic issues cards backed by stablecoin accounts, according to PYMNTS, putting a stablecoin funding source behind a conventional card credential.

How would networks and issuers verify intent?

Gandhi pointed to Visa’s Intelligent Commerce and Mastercard’s Agent Pay as signs that card networks may build on tokenization rather than create entirely new systems. Tokens are already used across merchants, issuers and acquirers as a way to represent payment credentials without exposing the original account details.

The next requirement, Gandhi said, is not limited to proving that an agent is authenticated. Payment credentials may also need to carry intent, such as whether a purchase matches a preapproved supplier, use case, price range or replenishment event.

That could create new questions over what transaction context is shared among networks, merchants, issuers and acquirers. PYMNTS reported that merchants may resist sharing line-item or proprietary information unless doing so improves approvals, lowers fraud exposure or reduces liability.

Lithic describes its Authorization Intelligence as a programmable governance layer for issuers. According to PYMNTS, the system can embed more complex rules, such as blocking spending when it is three standard deviations above historical patterns, rather than relying only on accept-or-decline decisions.

The larger issue for payments providers is how to move purchasing authority into software while preserving accountability and customer control. Gandhi’s view, as reported by PYMNTS, is that agentic commerce will depend on bounded, auditable transactions that merchants can accept without changing familiar payment flows.

This story draws on original reporting from PYMNTS.