Agent payments

What is Agent wallet (programmable wallet)?

Definition: An agent wallet is a programmable wallet built for an AI agent rather than a human: the agent gets its own payment credentials and identity, but every spend is constrained by human-set policies — spending caps, per-transaction limits and scoped permissions. Examples include Coinbase's CDP Agentic Wallets and Cloudflare Wallets.

Agents can't open bank accounts or click "Sign up with Google." Agent wallets solve this with programmable custody: keys held in secure infrastructure (MPC or enclaves), stablecoin balances, native x402 support for machine-to-machine payments, and guardrails the owner defines in code.

Two 2026 landmarks: Coinbase launched CDP Agentic Wallets (February 2026) — MPC-secured wallets with session caps, per-transaction limits and gasless settlement on Base, installable via CLI or an MCP server. Cloudflare announced Cloudflare Wallets with cloudflare.pay identity handles (August 4, 2026), giving agents deployed on Cloudflare a stable identity and human-set spending limits on x402 rails; handle reservations opened first, with wallet infrastructure rolling out over the following months.

How is an agent wallet different from a normal crypto wallet?

Policy is the product: a human owner sets spending caps, transaction limits and scopes, and the agent transacts autonomously only inside them. Keys stay in secure infrastructure (MPC/enclaves) rather than with the agent itself.

Do agent wallets only hold crypto?

Today they're mostly stablecoin wallets on x402 rails (e.g. USDC on Base or Solana), but the same pattern — scoped, delegated payment credentials — also exists on card rails via network tokens like Stripe's Shared Payment Tokens or Mastercard's Agentic Tokens.

Put it into practice

What an agent can buy with a wallet →

Related terms