Agentcard

Payments infrastructure for AI agents, enabling secure, user-approved online purchases.

Cover Block

Public sources

Field	Value
Name	Agentcard
Tagline	Payments infrastructure for AI agents, enabling secure, user-approved online purchases.
Headquarters	San Francisco, US
Founded	2026
Stage	Seed
Business Model	API / Developer Platform
Industry	Fintech
Technology	AI / Machine Learning
Geography	Global / Remote-First
Growth Profile	Venture Scale
Founding Team	Co-Founders (2)
Funding Label	Seed
Total Disclosed	~$8,000,000 [Agentcard, November 2026] [Fintech Global, November 2026]

Executive Summary

PUBLIC Agentcard is building payments infrastructure for AI agents, and it merits attention now because it paired a public product launch with an $8 million seed round in November 2026, backing a thesis that agent workflows may need their own payment and approval layer rather than a simple card-on-file workaround [Agentcard, November 2026] [Fintech Global, November 2026]. The company says it lets users save a card once, keep credentials hidden from the agent, and approve purchases through passkey, Face ID, or voice, positioning itself as middleware for agent-initiated commerce rather than a consumer wallet [Agentcard, November 2026] [Agentcard].

The founding story is early but legible. Karen Serfaty previously co-founded Atlas, an expense management startup acquired by Remote in January 2026, while Felipe Abello is described in public sources as Rappi’s first employee and former head of business operations, giving the team some proximity to both fintech operations and scaled commerce infrastructure [Remote, 2026-01-20] [FounderTrace].

Product scope is broader than a single card vault. Public materials describe four surfaces: Vault for storing and using a customer’s existing card without exposing card numbers to the agent, Issuing for virtual cards with budgets and controls, Voice for verbal approvals in early access, and opencommerce, launched on October 31, 2026, for agent-driven shopping across more than 5,000 stores [Agentcard] [AI Daily: Big Tech Free Cash Flow Goes Negative by $31.8B, 2026-10-31].

The differentiation claim rests on workflow depth, not just card issuance. Agentcard says it supports agent-browser integrations, MCP, CLI, and API access, and one third-party analysis reported sub-two-second checkout performance with a 99% success rate, though that operating claim appears lightly corroborated and should be treated as directional until independently validated [Y Combinator, November 2026] [AI Agents That Buy Mid-Conversation: Comparing the Payment Infrastructure Options].

The financing and pricing model are straightforward on paper. Agentcard announced an $8 million seed round with participation from Peak XV, Fin Capital, and Y Combinator, alongside angel backing from operators including Paul Graham, Adam D’Angelo, Tobi Lütke, Guillermo Rauch, Michael Grinich, Dharmesh Shah, Simón Borrero, and Darragh Buckley; its public pricing points to a mix of SaaS and take-rate economics, with Vault free up to 5,000 users, Issuing at $5,000 per month, and opencommerce charging 1% of completed orders on its standard tier [Agentcard, November 2026] [Fintech Global, November 2026] [Agentcard].

What matters over the next 12 to 18 months is not whether the product category sounds timely, but whether Agentcard can show repeatable enterprise adoption, durable approval workflows, and evidence that agent commerce extends beyond demos and founder-friendly design partners. Public source material still lacks named paying customers, revenue disclosure, or independent retention data, so the company looks promising but very early by the standards that matter for infrastructure investing [Fintech Global, November 2026] [Y Combinator, November 2026].

Lightly corroborated -- This section relies on company materials and one independent funding report, with partial corroboration from Remote and other public profiles.

Taxonomy Snapshot

Axis	Value
Stage	Seed
Business Model	API / Developer Platform
Industry / Vertical	Fintech
Technology Type	AI / Machine Learning
Geography	Global / Remote-First
Growth Profile	Venture Scale
Founding Team	Co-Founders (2)
Funding	Seed, total disclosed about $8,000,000

How the Company Got Here

PUBLIC

Agentcard appeared publicly in 2026 with a narrow proposition that fits the current agent software cycle: payments infrastructure that lets AI agents complete online purchases without exposing a user’s card details to the agent itself [Agentcard, November 2026]. The company is based in San Francisco, was founded in 2026, and presents itself as a platform for teams building agents rather than a direct-to-consumer wallet product [Agentcard, November 2026].

The verifiable timeline is short but coherent. Agentcard’s website shows a November 2026 launch period tied to the public announcement of an $8 million seed financing from Peak XV, Fin Capital, and Y Combinator [Agentcard, November 2026]. Around the same window, the company described four product surfaces, Vault, Issuing, Voice, and opencommerce, and positioned them as the core stack for agent payments and purchasing workflows [Agentcard, November 2026].

The founder record is directionally relevant, though the company website is the main source in scope for this section. Agentcard identifies Karen Serfaty and Felipe Abello as co-founders, and Y Combinator’s company profile describes the business as "cards for AI agents," which is consistent with the company’s own framing of secure, user-approved agent commerce [Y Combinator, November 2026]. No legal entity details or state filing information were confirmed in the materials provided, so the public record here rests mainly on company disclosures and Y Combinator’s profile page [Agentcard, November 2026] [Y Combinator, November 2026].

Lightly corroborated -- This section relies primarily on company website disclosures, with partial corroboration from Y Combinator.

Product and Technology

Sources and analysis

The product story is straightforward: Agentcard is trying to give AI agents a way to complete online purchases without exposing the underlying payment credential to the agent itself. On the company homepage and launch materials, the stack is presented as four surfaces: Vault, which stores a user or company card for agent-initiated purchases; Issuing, which creates one-time or reusable virtual cards with budget controls; Voice, an early-access approval flow; and opencommerce, a tool for searching, buying, and tracking orders across online stores [Agentcard, November 2026] [Y Combinator, November 2026] [Fintech Global, November 2026]. The framing matters because it places Agentcard less as a consumer wallet and more as middleware for teams building assistants, shopping agents, and internal enterprise agents [Agentcard, November 2026] [Y Combinator, November 2026].

The more interesting detail is where the approval and checkout logic appears to sit. Agentcard says purchases can be approved with a passkey, Face ID, and, in early access, voice, while Vault keeps the real card number away from the agent [Agentcard, November 2026]. The company and Y Combinator launch pages also describe support for native agent-browser integration, customer browser flows, MCP, a CLI, and an API, which suggests the product is designed to plug into several agent architectures rather than one prescribed interface [Y Combinator, November 2026] [Agentcard, November 2026]. Public customer examples on the homepage, including Orchid, Almanac, and Laso, are presented as usage illustrations rather than independently verified deployments, so they are best read as evidence of intended use cases, not proof of scaled production adoption [Agentcard, November 2026].

A few performance and pricing claims are public, but they should be handled carefully because corroboration is thin. Agentcard-related materials claim a 99% checkout-completion or success rate and sub-2-second merchant checkout for the Purchase Agent, while opencommerce has been described as covering more than 5,000 stores and was launched on October 31, 2026 [Y Combinator, November 2026] [AI Daily, October 2026] [AI Agents That Buy Mid-Conversation: Comparing the Payment Infrastructure Options]. Separately, the pricing page describes Vault as free up to 5,000 users, Issuing at $5,000 per month, Purchase Agent as free during beta, and opencommerce Standard at 1% of each completed order [Agentcard] [Pricing · Agentcard]. Those details make the commercial model legible, but most of the operating evidence remains company-originated, so the current read is product-complete enough to evaluate conceptually, not yet independently validated at scale.

Company-stated, unverified -- This section relies materially on company website materials and Y Combinator launch pages, with limited independent corroboration from Fintech Global and AI Daily.

Where the Demand Sits

PUBLIC The market matters now because the operational question around AI agents has shifted from whether they can reason to whether they can complete real transactions safely, and the public evidence around Agentcard sits squarely in that transition.

There is no source-grounded TAM, SAM, or SOM for Agentcard in the supplied materials, so sizing has to be handled cautiously and by analogy rather than assertion. The closest public demand signal is not payments volume, but agent adoption inside organizations: Salesforce said the average number of AI agents nearly tripled across organizations between February 2025 and April 2026 [Channel Dive]. That does not establish spend throughput for commerce or payments, but it does suggest a larger installed base of software agents that may eventually need authorization, budget controls, and payment rails if they move from answering questions to taking actions.

The demand drivers visible in the record are practical rather than thematic. Agentcard is positioned for teams building agents that need to buy online with a user-approved card, create funded virtual cards, or route purchases across merchants, according to the company homepage, pricing pages, and launch materials [Agentcard, November 2026] [Agentcard] [Y Combinator, November 2026]. The clearest tailwind is that the company is not asking the market to adopt a new consumer payment behavior first. It is attaching to existing card rails and existing e-commerce checkouts, while promising control layers that enterprises and consumer apps would need before letting an agent spend money on a user or company's behalf [Agentcard, November 2026] [Fintech Global, November 2026].

The nearest adjacent markets are card issuing, merchant checkout infrastructure, and enterprise spend management. That adjacency matters because Agentcard appears to be combining pieces that already exist separately: card vaulting for stored credentials, virtual card issuance with budgets, and a software layer that lets agents initiate purchases without directly handling the underlying card details [Agentcard] [Y Combinator, November 2026]. Karen Serfaty's prior link to Atlas, which Remote acquired in January 2026, is relevant here because Atlas operated in expense management for global teams rather than in consumer checkout, placing the founding context closer to controlled business spending than to pure shopping interfaces [Remote, January 2026] [TechCrunch, February 2024].

Regulation and macro forces cut both ways. The helpful side is that AI deployment continues to broaden, which increases the number of workflows where an agent might be asked to procure software, supplies, travel, or consumer goods on a user's behalf [Channel Dive]. The constraining side is that payments, KYC, delegated authorization, and card credential handling are not markets where product velocity alone settles the issue. Agentcard's own materials note that issuing requires KYC before the first card, which underlines that the category will be shaped by compliance obligations and fraud controls as much as by model quality or agent UX [Agentcard].

Cited market signal	Figure	Relevance
Average number of AI agents across organizations, Feb. 2025 to Apr. 2026	Nearly tripled	Indicates growth in potential software agents that may need payment or purchasing infrastructure [Channel Dive]
opencommerce listed stores, Oct. 31 2026	5,026 stores	Suggests broad merchant surface area for agent-led shopping, though this is product coverage rather than market size [AI Daily, October 2026]

The table does not size Agentcard's market directly, but it does frame the two public facts that matter most: agent deployment is rising, and at least one commerce-facing product from the company launched with multi-thousand merchant coverage. The gap between those facts and durable spend volume is still the main diligence question.

Lightly corroborated -- Section relies on one independent adoption signal from Channel Dive and one independent product coverage reference from AI Daily, with additional company and Y Combinator materials used for positioning rather than third-party market sizing.

Competitive Landscape

MIXED Agentcard is positioning itself less as a general fintech API and more as a control layer for agent-initiated commerce, which places it between card-issuing infrastructure, browser automation tools, and the emerging set of AI agent payment rails described in company and YC launch materials [Agentcard, November 2026] [Y Combinator, November 2026].

The competitive map is still forming, and the public record here is thinner than the product ambition. On one side sit incumbent payments and card infrastructure providers that already own merchant acceptance, underwriting, or developer trust at scale, even if the cited materials do not show them offering Agentcard's exact approval workflow for autonomous purchases [Agentcard, November 2026] [Fintech Global, November 2026]. On another side are early agent-payment products and commerce orchestration tools referenced in Agentcard's own comparison and explainer pages, which indicate that buyers are already evaluating alternatives across stored-card vaulting, one-time virtual cards, and purchase routing for agents, although the supplied research does not name enough of those vendors to support a clean public table here [Agentcard] [Y Combinator, November 2026]. A third bucket is adjacent substitutes: teams can stitch together browser agents, standard virtual cards, and manual approval flows internally, which may be less elegant but can be good enough for early deployments if transaction volume stays modest [Y Combinator, November 2026] [Agentcard].

Where Agentcard appears strongest today is product focus. The company is presenting a fairly opinionated stack, Vault for storing a user's existing card with approval controls, Issuing for budgeted virtual cards, Voice for verbal approvals, and opencommerce for search, buy, and order tracking across more than 5,000 stores, all under a single developer-facing brand [Agentcard, November 2026] [Agentcard] [AI Daily, October 2026]. That matters because the integration burden in agent commerce is unlikely to be just payments. It is approvals, checkout completion, and merchant coverage together. If Agentcard's reported sub-2-second checkout flow and 99% success rate hold up in third-party deployments, that would be a real edge, but those claims remain only partially corroborated in the materials provided and should be treated cautiously [AI Agents That Buy Mid-Conversation: Comparing the Payment Infrastructure Options] [Y Combinator, November 2026]. The durability of that edge is still uncertain: merchant coverage and workflow design can compound with usage, but checkout automation is also the kind of advantage that can erode if larger platforms decide the category is strategic.

The company is more exposed on distribution and trust than on product surface. Public sources do show a credible founder-investor network, including Peak XV, Fin Capital, Y Combinator, and several well-known angels, and that should help with introductions to agent builders and design partners [Agentcard, November 2026] [Fintech Global, November 2026]. Still, the cited record does not identify named paying customers, major enterprise deployments, or regulated moats beyond the product's stated KYC requirement for issuing, so Agentcard does not yet appear to own a channel that larger payment or software platforms could not contest [Agentcard] [Y Combinator, November 2026]. The homepage examples, Orchid, Almanac, and Laso, are useful proof of intended buyer segments, but they are presented by the company and do not by themselves establish depth of adoption or exclusivity [Agentcard, November 2026].

Over the next 18 months, the most plausible competitive outcome is not a winner-take-all market but a split between specialist infrastructure and broader platforms. Y Combinator is the named player most likely to look like a near-term winner if its portfolio flywheel keeps sending AI agent startups toward a default payments layer, because Agentcard is already using YC Launch as part of its go-to-market surface and benefits from that distribution signal [Y Combinator, November 2026]. Agentcard is the named player most likely to look like a loser if large buyers decide they prefer to assemble the stack from incumbent card and workflow vendors, since the current public evidence does not yet show customer captivity, exclusive merchant access, or a dataset moat that would make replacement costly [Y Combinator, November 2026] [Fintech Global, November 2026]. That leaves the core competitive question fairly plain: whether agent commerce becomes distinct enough to support a purpose-built vendor, or whether it gets absorbed into existing payments and automation rails before specialists can scale.

Opportunity

Upside Case

Public sources If Agentcard becomes the payment and checkout layer that AI agents call by default, the prize is not a niche fintech outcome but a control point inside a new transaction channel, where every agent-initiated purchase flows through a small set of trusted infrastructure providers [Agentcard, November 2026] [Y Combinator, November 2026].

The headline opportunity is straightforward: become the default infrastructure for AI agents that need to spend money online with human or company approval. That outcome is reachable, rather than purely aspirational, because Agentcard is not presenting a single feature so much as a stack. Its public product set spans Vault for storing a user card behind approval controls, Issuing for budgeted virtual cards, Voice for verbal approvals, and opencommerce for search, purchase, and order tracking across more than 5,000 stores according to public materials [Agentcard] [AI Daily, October 2026]. The company is also targeting developers and businesses building agents, not direct consumer adoption first, which matters because infrastructure companies usually win through integrations and repeat usage rather than brand marketing [Agentcard, November 2026] [Fintech Global, November 2026].

A second reason the upside deserves attention is timing. Salesforce said the average number of AI agents across organizations nearly tripled from February 2025 to April 2026, which suggests the agent layer is moving from experiment toward operational software [Channel Dive]. If that expansion continues, the need for spending controls, credential isolation, and merchant checkout tooling should rise with it. Agentcard's November 2026 seed round, with backing from Peak XV, Fin Capital, Y Combinator, and a roster of angel investors that includes Tobi Lütke, Adam D'Angelo, and Dharmesh Shah, gives the company credible early support as it tries to define that layer [Agentcard, November 2026] [Fintech Global, November 2026].

Scenario	What happens	Catalyst	Why it's plausible
Checkout rail for agent builders	Agentcard becomes the standard API that consumer and B2B agent companies embed when they need approved online purchasing	Broad adoption of agent products that need a reusable payment and checkout workflow	The company already offers API and browser-based tooling for businesses building agents, and publicly positions itself as infrastructure for consumer apps, B2B companies, and crypto teams [Agentcard, November 2026] [Agentcard]
Wallet and card operating system for enterprise agents	Enterprises use Vault for employee or customer cards, then expand into Issuing for policy-controlled agent spend	A category-tipping enterprise deployment where one team starts with approvals and expands into budgeted cards	The product line already separates user-card vaulting from company-funded issuing, which creates a natural expansion path inside one account [Agentcard] [Y Combinator, November 2026]
Commerce abstraction layer for agent shopping	opencommerce becomes a default way for agents to search and buy across fragmented online merchants without custom merchant integrations each time	Launch and scaling of the opencommerce layer across thousands of stores	Agentcard publicly launched opencommerce on October 31, 2026, and public references cite 5,026 stores, which at minimum shows an early attempt to aggregate merchant coverage into one interface [AI Daily, October 2026] [Agentcard, November 2026]

The compounding mechanism here is a classic infrastructure flywheel, if the early product claims hold. More agent builders integrating Agentcard should produce more transaction volume, which should improve checkout coverage, merchant edge-case handling, approval workflows, and product credibility for the next developer cohort. The public website already shows the company bundling merchant checkout, stored-card approvals, virtual card issuance, and a multi-store commerce layer into one system, and that breadth can matter because developers generally prefer fewer payment surfaces, fewer compliance handoffs, and fewer points of failure in checkout flows [Agentcard] [Y Combinator, November 2026]. If opencommerce usage grows alongside Vault or Issuing adoption, Agentcard could also gain a distribution advantage, because the team would sit at both the authorization layer and the purchase-execution layer.

The size of the win is still better framed through scenario logic than precise market sizing, because no credible public TAM figure was supplied in the source set. Even so, there is a concrete way to think about the outcome. Shopify, one of the clearest examples of a software company that became a transaction-layer control point, had a market capitalization in the tens of billions of dollars in 2026 according to public market data, though it serves merchants rather than agent builders [CompaniesMarketCap]. Agentcard is not Shopify, and the models are different, but the analogy is useful: if a company owns the trusted layer where a new class of commerce is authorized and executed, value can accrue quickly. Under the "checkout rail for agent builders" scenario, Agentcard could plausibly support a multibillion-dollar enterprise value over time (scenario, not a forecast) if agent commerce becomes a meaningful share of online purchases and if Agentcard captures both software revenue and a take rate on transaction flow [Agentcard] [Fintech Global, November 2026].

Lightly corroborated -- This section relies on a mix of company materials and one independent financing report, with additional scenario framing supported by public launch pages and trade coverage [Agentcard, November 2026] [Fintech Global, November 2026] [Y Combinator, November 2026].

Sources