Our customers are increasingly exploring the integration of AI-powered workflows with payment capabilities - what the industry now calls agentic payments. Lithic's virtual cards are designed with the intelligence and security controls needed for these autonomous spending decisions, making this new frontier both accessible and secure. For those unfamiliar with agentic workflows, we are talking about using large language models (LLMs) to create workflows where AI can move through multiple tasks like an “agent” by considering a number of complex factors to arrive at a comprehensive result. Agentic payments are the final leg of that journey to act on that result, and we wanted to use this forum to engage with innovators who have built these workflows already and want to use payments to close the loop.
The opportunity for agentic payments
Many of us have now experienced an agentic workflow through tools like Perplexity by asking for dinner recommendations with a complex prompt like: “Find me an Italian restaurant in Soho NYC that has a Google review rating higher than 4.7 with outdoor seating and has more than 2 vegetarian entrees on the menu.” Agentic payments are the next logical step where the agent actually follows through on a request like: “Place a DoorDash order for 2 vegetarian entrees and a cheese pizza from an Italian restaurant in Soho NYC that has a Google review rating higher than 4.7 and have it delivered at home at 6:30PM.”

Agentic workflows that include an agent making a payment are clearly not limited to just ordering food. We are already seeing workflows that optimize travel bookings, ad spend management, bill payments and light procurement. Adding a payment mechanism to these flows is simply a natural extension of where we and others focused on this topic see things moving.
We see this becoming a game changer for procurement use cases. As an example, a marketing team asks an AI agent to allocate $10k across multiple ad platforms (Google, LinkedIn, etc.). The agent analyzes the target audience, historical performance data and ad rates, and then decides to allocate the $10k ad spend budget across the various platforms against real-time pricing. The agent spins up a Lithic virtual card for each platform with spending limits using Lithic’s authorization rules and executes ad spend buys autonomously. In this example, agentic payments have created both efficiency and cost savings by buying ads in real time, optimizing marketing performance.
Example code demonstrating how Lithic can enable an AI agent to restrict a card to only spend at ad agencies by amount per day/week/month

Ultimately, agentic workflows resemble expense management workflows, where employers are enabling their employees to spend company money. Instead of a single employee like a CFO or a few senior leaders making every single purchase decision, next-generation spend management tools are already creating efficiency by pushing decision-making out to a trusted employee base while embedding controls to ensure that the company is not putting itself at risk. When we think about agentic payments, we are simply talking about putting a controlled payment mechanism like an issued card into the metaphorical hands of an AI agent, instead of an employee.

What controls look like at the speed of AI
To make this work from a regulatory perspective, we need to know who is directing the agent to make spend decisions (KYB/KYC) to ensure that we’ve considered AML and sanctions screening obligations. This could start without a rigorous approach, but as more funds move there will be a graduation through a tiered model similar to what we see with Square, Stripe and others. From a control perspective, we will want to make sure that cards have tight controls that can only be used for specific transactions based on strict parameters of who, when, and where spend can take place. These types of controls have been Lithic’s bread and butter for more than a decade.
Our journey with the Lithic platform began nearly a decade ago when we migrated from our previous card processor. Initially launched as Privacy.com, we created a solution for individuals and businesses to generate virtual cards specifically designed to protect their primary funding sources.
This addresses the same fundamental concern people have about agentic payments today: what safeguards exist when AI agents gain access to our bank accounts and potentially spend in unintended ways?
What we were missing a decade ago from the existing providers in the space was the ability to balance complex rule sets to protect cards while also ensuring a high level of reliability. Over the last decade, we have built infrastructure that is architected to be flexible and resilient enabling our customers to make targeted decisions on card transactions as they happen. We are introducing programmable code to enable complex rules against individual cards, and we’re doing it in a way that doesn’t degrade speed. For more details on our unique architecture, I would encourage you to read our post on Building an Authorization Rules Engine.

Lithic is uniquely positioned in the market to help innovators add card issuing to their agentic workflows as a result of our architecture. In broad strokes, we can enable these new agentic offerings to quickly onboard onto our platform based on what we do for countless businesses who issue virtual cards on their own behalf, like we do for our customers on Privacy.com, as well as for clients like Order, Medallion, Torch, and Gordian.
In all of these examples, we are already issuing controlled virtual cards against a primary funding source that can be locked to a merchant for a limited duration and up to a certain amount. We recognize that LLMs are enabling us to do magical things already that help us make better decisions and improve efficiency. We believe that adding payments is the next logical step, and it is just a matter of providing the right level of controls to give us confidence in the agent's actions.
How Lithic can help mitigate risks while enabling agents
In order to make agentic payments using cards work, there are a few key components that will be required to ensure the safety of funds while also harnessing the efficiency that these workflows bring. These include onboarding, complex authorization controls, fraud rules, transaction monitoring, and dispute management.
Onboarding
Everyone (business or individual) needs to set up an account so that we can ensure that no nefarious actors are getting access to financial networks. Once we have an account, agentic workflows are just another way for someone to leverage their account. The best place to get started on understanding these flows is through our reference guide focused on setting up new accounts. This guide walks through the entire process of opening a new account and validating ownership which can be done using our API toolset or as a fully managed offering.
Fraud rules
In addition to authorization controls that are tied to specific cards, most entities will also want to have baseline rules established that enable them to keep cards from being abused. Through our Authorization Rules toolset, we have made it simple for customers to build rules that stop transactions before they even hit our authorization controls. In addition, we have also developed a set of rules based on Lithic’s years of experience that run in the background providing another safety net from large scale attacks or errors in execution.

Complex authorization controls
Since virtual cards can be spun up for any transaction, you will want to include complex parameters that limit the usage of that card. Through our Auth Stream Access (ASA) feature, we can enable customers to build unique rules for each card. In the Doordash example from earlier, the card could be restricted only to a specific restaurant to be used for no more than $200 on a specific day and a specific time period. Over time, we will continue to provide more ways to program code that enables those directing AI agents to build additional methods for card security.
Exception handling - disputes and chargebacks
Inevitably, there will be situations where an agent will make a purchase for a product or service, and a dispute needs to be filed. Providing the right tools and data to file those disputes will be critical to any large-scale deployment of agentic payments. We expect that agentic workflows will enable our customers to file and manage disputes quickly via an API-based dispute solution. For more complex disputes, customers can always use our managed dispute service to resolve harder issues.

At Lithic, we are committed to building a tech-centric core processing engine for issuing cards that is future-proofed. We’re already building out ways to streamline interactions with Lithic using a forthcoming MCP, and we’re excited about how our customers can leverage this as a starting block to build upon our tech stack. If you are an innovator in the agentic workflow space, we encourage you to read through our API documentation and engage with our team.
You can integrate agentic payments into your workflows today with the Lithic platform. Let's build something remarkable together!