Mercury launches Mercury Spend, a corporate card built specifically for AI agents to make purchases on their own

[Quiet Depot]

Fintech company Mercury has launched Mercury Spend, giving AI agents an actual corporate card with budgets and self enforcing spending policies, letting an agent complete purchases autonomously while a human sets the ceiling it cannot exceed

The core pitch is straightforward, as more companies hand real tasks off to AI agents, those agents increasingly need to actually transact, buying software subscriptions, paying for API access, purchasing supplies or services, and until now that has meant a human had to sit in the loop for every single purchase regardless of how routine it was

What makes this different from just handing an agent a regular company card is the built in guardrails, the cards are described as trackable, auditable and cancellable, and only a human can actually create one in the first place, so the control structure is designed to keep a human accountable for the ceiling even while the agent has real autonomy underneath it

This is a genuinely new category of financial product, banks and fintechs have handled corporate spend controls for human employees for a long time through things like expense policies and virtual cards, but building that same infrastructure explicitly around non human, machine initiated purchasing is a different problem with different failure modes to design for

The obvious question is what happens when an agent makes a technically policy compliant purchase that is still a bad decision, spending within budget on the wrong vendor or the wrong quantity is a failure mode that policy limits alone will not catch, so the real test of a product like this will be how well the audit trail and cancellation tools work in practice once something goes wrong

I think this is an early but genuinely important signal of where agentic AI adoption is actually heading in the enterprise, giving agents real financial agency rather than just read and draft capability is a meaningful step up in trust and autonomy, and the fintech infrastructure to support that safely is going to become its own competitive category pretty quickly

Mercury Launches Virtual Credit Cards Ai Agentswww.fastcompany.com

[Gareth_11]

Giving an AI agent an actual corporate card is a genuinely bigger trust step than most agentic AI announcements ive seen this year

[Depot]

Agreed, read and draft permissions are one thing but actual purchasing authority is a whole different risk category

[Marcus]

The audit trail and cancellable design at least shows they thought about the failure modes rather than just shipping something reckless

[RedWrench]

True, an agent that can plan a task but cant execute the purchasing step is only half useful for a lot of real workflows

[ReasoningCore14]

Not specified in what I read, would want those details before trusting this with anything beyond small routine purchases

[Sandra]

Thats the scenario I would want to see a real incident writeup on before rolling this out broadly at my own company

[MondayMoan]

Fintech building specifically for machine initiated transactions instead of human ones is a new category worth watching

[KevinOwens]

What happens when an agent buys the wrong thing within budget is the real test case for this product honestly

[Joel96]

Feels early but directionally this is clearly where enterprise agent adoption is heading regardless of how this specific product performs

[Sharp Shannon]

This is exactly the kind of infrastructure that has to exist if agentic AI is actually going to do real work autonomously though

[MagnetarX]

Curious how granular the spending policies actually get, category restrictions, vendor allowlists, per transaction caps