How will my agent pay for things?

In the not-so-distant future, consumers and businesses will be able to instruct an agent to make any payments they want on their behalf.

Customers will be able to book a group holiday — including all flights, hotels, and activities — with a single click, as their agent finds the cheapest, most points-optimized itinerary tailored to their unique preferences. Likewise, businesses will be able to monitor inventory and have agents automatically make the purchases to restock.

Indeed, early efficiencies are already live in the market and are beginning to accelerate with browsers, PSPs, the networks, and more racing to secure their place in the agentic commerce future. In the last month alone, both Mastercard and Visa launched AI agent payment solutions, while PayPal introduced its first MCP server. Through PayPal’s MCP, LLMs can now instantly generate invoices and share client-ready payment links on behalf of merchants — eliminating the need for manual invoice creation or direct integration with PayPal’s APIs.

However, agents today are not yet truly autonomous, particularly when it comes to handling payments.

What needs to be solved for agents to actually pay for things?

1. Role and Scope of the Agent

Is there a human or business “linked” to this agent? What is the allowable purchase scope? (Which items, from which merchants, for how much?)

There is currently no agreed-upon standard for how a human assigns a role and scope to an agent, nor for how a merchant verifies that role and scope. For example, it’s unclear how a person can designate whether an agent is “attached” to them, what the agent is allowed to do, or how much it is authorized to spend. On the other side, merchants have no consistent way to verify whether they’re interacting with an agent versus a human, and if it is an agent, what permissions or limits have been granted.

2. Fraud and KYA (Know Your Agent)

Is this an authorized AI agent?

Scammers stole over $1 trillion globally in 2023. Without sufficient safeguards, that number could grow exponentially with AI. Sardine AI, for instance, has already identified fraud targeting agentic commerce and is actively protecting its clients against increasingly sophisticated techniques, including the use of AI-powered voice modifiers by attackers.

3. Liability

Who is responsible and who bears the cost?

Chargebacks, for example, raise a fundamental question of liability. When a transaction is reversed, someone must absorb the cost. Yet, with an additional party involved in agentic payments (the agent!), allocation of responsibility is unclear.

Agentic commerce opens a myriad of opportunities for new entrants. For example, as AI agents increasingly transact with each other, they may need to execute agent-to-agent payments or micro-transactions for data access. Stablecoins, with their cost-effectiveness and programmable capabilities, could be well-suited for these machine-to-machine transactions.

For a quick preview of exciting companies in the space:

Paid.ai is making sure businesses bill accurately to monetize the full value agents deliver.Nekuda is building payment rails designed specifically for AI agents.Payman is arming every AI agent with its own wallet so it can securely move money.

We’ll follow up soon with a deeper analysis of this rapidly evolving space. If you’re building in this area or have insights to share, we’d love to hear from you.

More from the a16z Fintech Team

On the Wharton FinTech Podcast, a16z partner Melissa Wasser discusses how fintech founders can evaluate M&A opportunities, structure smarter capital strategies, and assess the tradeoffs between equity and debt financing. Listen now »

Financial services are famously human capital-intensive; even at Goldman Sachs there’s still a “Federation” of back office workers sitting in Excel, says a16z general partner David Haber. On TPBN, he argues that much of that work — across legal, compliance, risk assessment, vendor onboarding, and more — should be handled by AI. Watch now »

Ben Franklin famously remarked there are only two certainties in life: death and taxes. a16z partner Marc Andrusko makes the case that AI should be added to that list. On the Wharton FinTech Podcast, he unpacks the ways AI can streamline workflows and automate core tasks in finance and accounting. Listen now »

Recent M&A Deals and Market Intel

eToro opened for trading on May 14 after pricing nearly 12 million shares at $52 each — above its initial target of 10 million shares between $46 to $50. The upsized deal was driven by strong demand, with an order book reportedly 20 times oversubscribed, according to Bloomberg.

Chime filed its S-1 on May 13 and is reportedly seeking a valuation between $8 billion and $9 billion.

Klarna is now expected to delay its IPO until this fall due to ongoing market uncertainty, according to Realtid, a Swedish news source.

Monzo has reportedly lined up Morgan Stanley to pitch the company to investors in preparation for a potential IPO in the first half of next year.

iCapital announced its acquisition of Citi Wealth’s alternative investments feeder platform on May 13. Leveraging its technology, iCapital will streamline the operations and management of Citi’s existing and future alternative investment funds platform. Additionally, it will strengthen Citi Wealth’s global sales capabilities through a dedicated team of alternative investment specialists.

AvidXchange announced its take-private sale to TPG and Corpay in a cash transaction that values the company at $2.2 billion. The deal reflects valuation multiples of approximately 4.9x LTM revenue and 25.6x LTM EBITDA. According to the agreement, Corpay will invest $500 million in exchange for a 33% ownership stake.

Mastercard and Corpay launched a strategic partnership in cross-border payments on April 29, with Mastercard investing $300 million for an approximately 3% equity stake in Corpay’s cross-border business. The investment values the business unit at an enterprise valuation of $10.7 billion, and represents a forward EBITDA multiple of 20x.