AI Agents Need Permissioned Funding, Not Bank Accounts

After all, if a multinational eventually operates, say, 10,000 artificial intelligence agents buying software, booking travel, procuring inventory, managing advertising and negotiating with suppliers; then funding 10,000 separate accounts would reproduce one of corporate finance’s oldest problems at machine scale: cash sitting in thousands of places waiting to be used.

Forward-thinking CFOs are already hoping this age-old dilemma can be avoided by financial architecture that separates money from permission to spend it. The question for finance teams is not where an AI agent’s money lives but who has authorized the agent to create a financial obligation, how large that obligation can be, and which pool of liquidity should fund it once the transaction is approved.

Read more: B2B Software’s Next Competitive Moat Is Permission, Not UX 

The Corporate Agentic Wallet May Not Look Like a Wallet

Corporate payments have historically bundled several functions into a single product.

A corporate card identifies the spender, provides access to credit, imposes spending controls and supplies a payment credential. A bank account simultaneously stores money and provides the mechanism for moving it.

AI agents give finance an opportunity to unbundle those functions. Visa and Mastercard’s emerging agentic-payment architectures, for example, already point toward credentials, authenticated intent and programmable controls rather than simply handing autonomous software conventional payment accounts.

And the virtual cards form payment networks may actually be the most obvious early agentic winner because they already approximate what AI agents need. Virtual cards can be generated programmatically, restricted to particular merchants, limited by transaction amount or time period and reconciled against specific purchases. The underlying capital can remain with the issuer until a transaction occurs.

More importantly, cards connect autonomous software to a merchant acceptance network that already exists. An agent buying airfare, advertising or office supplies does not need a revolutionary settlement rail if the merchant already accepts cards.

The PYMNTS Intelligence report “How Agentic AI Went From Zero to CFO Test Runs in 90 Days” found in September 2025 that nearly 7% of enterprise CFOs in the United States had deployed agentic AI in live finance workflows, while an additional 5% were running pilots.

See also: AI Agents Push CFOs to Rethink Business Payments

We’d love to be your preferred source for news.

Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!

Stablecoins Add Value Where Machines Are Paying Other Machines

Stablecoins become more compelling in a different environment. Imagine an AI agent purchasing compute every few seconds, paying for API calls, acquiring data, licensing content or compensating another autonomous service. The transaction could be global, continuous and extremely small. That is not necessarily what card infrastructure was economically designed to do, but it does represent one of the value propositions the proponents of on-chain digital dollars are clamoring loudest for.

But even here, the idea that every agent needs its own pile of stablecoins is questionable.

A more capital-efficient model resembles corporate cash concentration. Treasury maintains a central stablecoin balance while individual agents receive virtual wallets or sub-ledgers representing claims against it. Policy engines determine which agents can access which amounts, counterparties and transaction types.

The important innovation is again the concept of centralized liquidity paired with decentralized authority. That architecture could eventually apply whether the underlying asset is a stablecoin, tokenized deposit or conventional bank money.

Read more: How $100 Million CFOs Are Setting Their Neocloud Budgets 

The most disruptive possibility of agentic funding mechanisms, however, sits one layer deeper. Agents may not need prefunded money because they could obtain purchasing power at the moment they need it. The authorization to buy and the authorization to borrow are already beginning to converge, with Visa on Tuesday (Sept. 8) announcing a new credit and funding product for its stablecoin-backed card suite.

Instead of asking whether Agent 4382 has $100,000 of available credit, a bank could decide whether Agent 4382 should be permitted to create an $18,700 liability for a particular purchase at this particular moment.

Still, the PYMNTS Intelligence report “Tech on Tech: How the Technology Sector Is Powering Agentic AI Adoption” found a widening agentic readiness gap between tech companies and firms in goods and services, with 75% of tech firms reporting they were extremely familiar with agentic AI, versus 33% of goods firms and 38% of services firms.

For all PYMNTS artificial intelligence, B2B and digital transformation coverage, subscribe to the daily AI, B2B and Digital Transformation Newsletters.