AI agents can now use virtual cards to pay suppliers, book travel and buy supplies, without an employee entering payment details for every purchase. The interesting shift is not that a bot has acquired a wallet, but that businesses are starting to let software spend within rules they set.
Watch Desk analysis
What happened
PYMNTS reports that Ramp, Mercury, Corpay and Sunrate are among the companies introducing cards or payment tools for autonomous agents. Their approaches vary: Ramp can issue a single-use card tied to one agent and transaction, while Mercury lets a person set a budget, merchants and spending categories. Sunrate’s virtual Agentic Card can be set up for a particular workflow, with limits on amount, currency, merchant category, region and expiry.
The common thread is controlled access rather than a blank cheque. Businesses set spending rules; the agent can make an authorised purchase, but cannot raise its own limit. PYMNTS also describes consumer payment products from Stripe and Crossmint that let agents pay using an existing card or wallet arrangement.
Why it matters
Giving software permission to act is one thing. Giving it a way to spend makes the consequences rather more tangible. Single-use cards and transaction caps could let companies automate routine purchases without handing an agent unrestricted access to an account.
The gap between interest and adoption is still notable. PYMNTS Intelligence says 52% of CFOs surveyed would use autonomous AI to recommend changes to liquidity or payment timing, but fewer than one in three would let it reallocate budgets. The report says its survey covered 60 US company CFOs in October 2025. A separate PYMNTS Intelligence report from September 2025 found close to 7% of enterprise CFOs had deployed agentic AI in finance workflows, with 5% running pilots. These are survey findings, not a count of all companies currently using agent payments.
Our read
This is a practical step towards agents doing useful work beyond clicking through a website. The sensible design pattern, at least on paper, is narrow authority: a human approves the job, the payment instrument is constrained, and the transaction can be checked afterwards. That is less cinematic than “AI takes over finance”, and considerably more reassuring.
The hard question is what happens when an agent makes a permitted payment for the wrong reason, or a merchant and workflow are misidentified. Limits can restrict the damage, but they do not settle who is accountable when the agent gets it wrong. Treat these products as new tools for controlled automation, not evidence that finance teams are ready to hand over the keys.
What to watch
- Whether businesses move from pilots and limited workflows to routine agent-initiated payments.
- How card providers handle disputes, fraud and mistakes made within an approved spending limit.
- Whether companies can inspect which agent initiated a transaction and what instruction it was following.
Discussion spark: Would you let an AI agent make purchases with a tightly capped, single-use card, or should every payment still require a person to approve it?
Sources and evidence
- Source update (5 November 2026, 18:47 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.