Mercury Gives AI Agents a Credit Card, and That Changes the Architecture of Corporate Spending
The average founding team of a startup used to have two people and three engineers. Today it can include a dozen artificial intelligence agents completing tasks in parallel, negotiating prices with vendors, or purchasing software without any human approving each individual transaction. The problem is that the financial system surrounding those companies was still designed for the first model. Mercury has just begun to change that.
The digital banking company, founded in 2019 and turned into the favorite bank of Silicon Valley startups, has just launched what it calls AI agent cards: virtual credit cards that companies can issue directly to their autonomous systems so they can make purchases without human intervention at every step. The launch is not a speculative pivot toward the future. It is the formalization of something that was already happening without adequate infrastructure.
Immad Akhund, co-founder and CEO of Mercury, was direct about it: customers were already manually issuing virtual cards to their agents. What was missing was the control framework, the audit trail, and the programmable limits that turn that informal practice into something a CFO can defend in front of a board of directors. That is exactly what Mercury is building.
Autonomous Spending as an Infrastructure Problem, Not a Philosophical One
There is an enormous distance between saying that AI agents are going to change work and actually building the financial plumbing to make that function in practice. Mercury is betting that this gap is where the business opportunity lies.
The logic behind agent cards is not complicated: if a company delegates to an agent the task of managing software subscriptions, optimizing advertising campaigns, or purchasing services on behalf of the team, that agent needs payment capability. And that capability needs to have clear limits, auditable records, and the ability to be revoked in seconds. Without that, the alternative is for a human to approve each transaction individually, which eliminates precisely the benefit of having autonomous agents.
The architecture that Mercury is proposing recognizes that credit cards already have something that no alternative payment system has managed to replicate at scale: universal acceptance, fraud protection, dispute mechanisms, and decades of built infrastructure. Akhund points this out with precision: for routine purchases below $2,000, cards are the most sensible practical tool available right now. Not because it is the perfect long-term solution, but because it is the one that works with the existing vendor ecosystem without asking anyone to change their billing system.
This is not a minor detail. The startups that are testing operating models with autonomous agents cannot wait for a completely new payment infrastructure to mature. They need tools that work within the rails that already exist. Mercury is responding to that concrete friction, not to a hypothetical scenario.
The move also has a clear retention logic. Mercury serves, according to its own data, one in three startups in the United States. Between 2024 and 2025, it doubled the number of customers in the AI startup segment. If those companies begin using autonomous agents for operational spending and Mercury does not have a specific answer for that use case, the risk is that they migrate to more specialized platforms. The agent cards are, in part, a retention mechanism disguised as a product innovation, and there is nothing wrong with that when the friction they resolve is genuine.
Mercury Against the Fintech Board and What It Reveals About the Market
The competitive context of this launch matters as much as the product itself. Mercury is playing on a board where Ramp was valued at $44 billion in June 2026 and where Capital One acquired Brex for $5.15 billion earlier in the same year. Both platforms built their reputations specifically in spend management and corporate controls: the exact territory where Mercury had historically been weaker.
Mercury came to market with a different proposition. Its original advantage was not spend management; it was the branch-free onboarding experience and affinity with software founders. That worked very well at the acquisition stage, especially during the collapse of Silicon Valley Bank in 2023, when it captured $2 billion in deposits and 8,700 new customers in just a few days. But the same customer base that arrived looking for a reliable bank account eventually needs more sophisticated spending controls. If Mercury does not have them, Ramp or any other specialized platform fills that space.
The spend management expansion that accompanies the launch of agent cards — which includes budgets, automated accounting, and policy enforcement — is the direct response to that risk. Mercury is attempting to become the complete financial layer for startups rather than just being their bank, and it is doing so precisely when the market for autonomous agent tools is early enough that it makes sense to position there.
Mercury's move must also be read in parallel with its national banking license application before the Office of the Comptroller of the Currency, which received conditional approval in April 2026. That license is not a minor regulatory detail. It is a strategic decision that reduces dependence on intermediary partner banks, accesses Federal Reserve payment rails directly, and structurally improves margins. The lesson of the Synapse collapse — where Mercury was exposed as the largest client of that intermediary platform when customer funds disappeared — is clearly incorporated into this decision.
A company that aspires to be the financial infrastructure for the next generation of AI startups cannot depend on intermediaries whose soundness it does not control. The banking license and the agent cards are two pieces of the same move: reducing friction at both ends of the business, operational and regulatory.
What Agents Reveal About How Real Work Is Being Reorganized
Beyond Mercury as a company, the launch of agent cards is a signal about how the operational structures of the newest startups are changing. Sam Altman spoke in 2024 about the possibility of a billion-dollar, single-person startup powered by AI agents. That image circulated as a futuristic provocation. Today it is beginning to have concrete financial consequences.
A company that operates with three people and twenty autonomous agents has radically different spending needs than a company of twenty people. The volume of transactions may be similar, but the distribution of who executes them changes completely. Approval workflows designed for humans — which assume that someone is going to review a request, understand the context, and click approve — make no sense when the agent needs to purchase a data API at 3 in the morning to complete an analysis before the human team starts its workday.
This is not a problem of trust in AI. It is a process design problem. The companies that are adopting autonomous agents most seriously are discovering that the friction is not in the agent itself, but in everything surrounding its ability to act: access to tools, system permissions, and now, payment capability. Mercury is attacking that final link in the chain.
What makes the case interesting is not that Mercury invented something technically impossible before. Virtual cards have existed for a long time. The novelty lies in recognizing that the use case has changed enough to merit a specific product configuration: per-agent limits, individual visibility of each expenditure, instant revocation, integration with the approval workflows of the rest of the platform. The technology is the same; what Mercury redesigned is the control model around that technology.
There is a genuine adoption question that this launch does not yet answer. Mercury acknowledges that the volume of agent transactions is small today. The market of startups with agents autonomous enough to require independent spending power is still a limited segment. The bet is that that segment will grow and that being positioned in the infrastructure from now has strategic value even if the incremental revenues are modest in the short term.
That is, ultimately, the correct reading of this move. It is not a product that solves a massive problem today. It is a positioning statement about where Mercury believes the flow of value will be when autonomous agents transition from being internal experiments to becoming a standard part of the operation of any software-intensive company. Whoever holds the spending infrastructure when that moment arrives will also hold the lever to capture the rest of the financial relationship with those companies. Mercury is buying that right now, before it becomes expensive.
