YC-backed Agentcard launches rewards for purchases made by AI agents

Why it matters

Agentcard is turning agent payments into a rewards loop, giving users and developers a financial reason to route more autonomous spend through its cards.

Karen Serfaty (@keyserfaty) said Friday that Agentcard is launching a rewards program that gives users tokens for every purchase made with an Agentcard, adding a loyalty layer to a product built to let AI agents spend money online without handling a user's primary card details.

Serfaty announced the program in a thread on X on July 17th, writing that traditional credit-card rewards were "built for a different time" and that Agentcard would give users "tokens" instead of points. The thread said the rewards program is live today and pointed users to agentcard.sh to implement it. Agentcard has not disclosed in the thread the token type, earning rate, redemption mechanics, transferability, expiration policy, or whether the rewards apply to all transaction types.

That missing detail matters because Agentcard is operating in a category where rewards are only one part of the trust problem. Agentcard's core product is card issuing for agents: its homepage pitches card issuing for agent-first startups, with one-time cards, human authorization, fraud controls, an MCP server, a CLI, and an API. The company's API page says developers can issue single-use virtual Visa cards with fixed spending limits, while the docs describe cards that AI agents can create, monitor, and close programmatically.

Serfaty is not coming at the problem as a first-time payments founder. Y Combinator's Agentcard profile lists her as CEO and founder, previously CEO of Atlas, an expense card for global teams that was acquired by Remote.com, and later GM of Cards at Remote.com. It also lists co-founder Felipe Abello (@PipeAbellos) as a prior co-founder of Infactory.ai, founder and CEO of Nara, and the first employee at Rappi. Agentcard is a Summer 2026 Y Combinator company based in San Francisco, according to YC's directory.

The rewards launch lands after several weeks of visible product hardening. Agentcard's updates page shows a July 10th dashboard launch for managing cards, wallets, activity, connections, organizational settings, webhooks, API credentials, and billing. On July 12th, Agentcard said organizations could automatically earn back a share of every transaction their cards make, credited into a company wallet. On July 13th, Agentcard added organization-set markup on checkout and wallet-funding transactions, in addition to the card-interchange share flowing into the company wallet.

That sequence explains the incentives under the rewards program. Agentcard is trying to become the spend layer for agents and agent-enabled applications. Rewards give users and developers a reason to route more transactions through Agentcard, while the company is also giving organizations a way to capture economics from transactions their integrations generate. The card is the control surface; the reward is the nudge to keep spend inside Agentcard's rails.

Agentcard's public materials also show how fast the product definition is moving. The YC profile describes Agentcard as "debit cards for AI agents." Agentcard's current terms, updated July 10th, describe a different legal shape: single-use virtual Visa cards designed for AI agents on behalf of a verified human account owner, structured as set-limit, fully collateralized credit accounts funded in advance from a wallet, issued under a card program by Third National. The docs also note a migration to stablecoin-collateralized Visa cards funded by a per-user USDC wallet and require users to complete or redo KYC before their first new card.

For Agentcard, that legal and operational plumbing is the harder part of the product. A conversational agent buying sushi, groceries, flights, or API credits is easy to demo. Making the spend scoped, funded, verified, cancellable, auditable, and economically useful to the developer integrating it is the actual company. Agentcard's own updates show that the hard edges are still being sanded down: recent fixes cover identity verification, webhook scoping, checkout estimates, refund behavior, cart state, OAuth, account duplication, and the difference between sandbox and live-money issuance.

The reward program puts a consumer-fintech mechanic on top of that infrastructure. It also sharpens a question Agentcard will have to answer directly as usage grows: whether the tokens are simply a usage credit inside Agentcard, a broader rewards asset, or something with regulatory and accounting consequences beyond loyalty. Agentcard can sell the product as a safer way to let agents pay. Once Agentcard starts rewarding every purchase with tokens, the reward design becomes part of the trust model too.