Analysis
a16z's new map splits agent payments into four rows. What each row means, where Prava sits, and five questions to ask anyone in delegated payments.

a16z just put Prava on its map of the agent economy, in the payments section, under delegated payments.
For about twenty years, websites have tried to keep bots out, with tests like clicking every square that contains a traffic light or typing distorted letters to prove you're human.
a16z's "Welcome to the B2A Economy", published on November 10, argues that this is changing. Its one-line summary reads: "Companies spent decades keeping bots out. Now they'll compete for their business."
Agent traffic is already growing
The article backs this up with numbers. a16z cites Human Security seeing agentic browser traffic grow 8x between August and October, with Meta's Muse driving 60% of it. It also cites Shopify: AI-driven traffic to its stores grew 8x year over year in Q1 2026, and orders from AI-powered searches grew nearly 13x.
To show who is building for this new kind of customer, a16z published a map of the companies involved. Prava's row is called delegated payments: products that let an agent spend a person's or a company's money, with their permission.

The Payments and Checkout section of a16z's agent economy infrastructure map. Prava is in the delegated payments row. See the full map in a16z's Welcome to the B2A Economy.
Below, we explain what each row in the payments section means, and why Prava also belongs in the checkout row.
The four rows in the payments section
Row on the map | What it means | Where Prava fits |
|---|---|---|
Delegated payments | An agent pays with someone else's money, with their permission | Prava is listed here: the user's own card, with a one-time network token for each purchase |
Agent accounts and spend controls | The agent holds its own account or wallet, with limits | Prava uses the user's card instead, so there's no balance to fund |
Identity and trust | Proving who the agent is and who it acts for | Each Prava payment carries a record of the user's approval, and agentic tokens tell the card network an agent acted with consent |
Checkout | Getting the order through the store | Prava covers this row too: product search across UCP stores and checkout in 5 to 20 seconds |
Three models within delegated payments
Products in the delegated payments row work in quite different ways. They fall into three models:
A new card for the agent. The provider pays the store with a card it issues, then charges the user or draws on a balance. Stripe's Link works this way for most stores (how Link compares with Prava).
The user's own card, kept in a vault. The provider stores the user's existing card and uses it within limits it sets in software.
A network token on the user's own card. The card network issues a token for one store, one amount and a short window, and enforces those limits itself. This is how Prava works.
These models lead to different results for the user: whose card the merchant charges, what the statement shows, and where a dispute goes. We compare two of them in Virtual cards vs network tokens for AI agents.
Where Prava fits
Prava uses the third model, and it covers two rows of the map. In delegated payments, the user adds their card once, approves each purchase with a passkey inside your app, and the merchant charges that card through a one-time network token from Visa Intelligent Commerce or Mastercard Agent Pay. In checkout, Prava's Shop API finds products across UCP (Universal Commerce Protocol) stores and merchants like Macy's, and completes checkout in 5 to 20 seconds on UCP stores.
Both come from one API, so an AI app doesn't have to stitch together a payments provider and a checkout provider. See how the pieces fit together.
Questions to ask a delegated payments provider
Whose card does the store actually charge?
What will the user see on their statement?
Where does a dispute go: the user's bank, or the provider's process?
Does the user need a new account, a new app or an identity check?
Who enforces the limits: the provider's software, or the card network?
The answers show which of the three models a provider uses.
Summary
a16z splits agent payments and checkout into four rows: delegated payments, agent accounts and spend controls, identity and trust, and checkout.
Delegated payments holds very different models, from new virtual cards to network tokens on the user's own card, so the label alone doesn't show how a product works.
Prava sits in delegated payments and also handles checkout: one API for product search, checkout and payment on the user's own card.
To try Prava, create sandbox keys in the dashboard and follow the quickstart.
Sources
a16z: Welcome to the B2A Economy, Olivia Moore and Justine Moore, Nov 10, 2026, and its infrastructure map

Prava team

