AI agents used to stop dead when they hit a paywalled tool: they could call a payment API someone else wired up, but they couldn't sign up for a new service mid-job. Machine-native payments (crypto wallets, stablecoins, protocols like x402) change that. Explore how each autonomy level changes how often a human must step in.
All figures are illustrative estimates for teaching, not benchmarks.
Every agent job is a loop: plan → act → hit a missing/paywalled tool → resolve → continue. The resolve step is where autonomy is decided. Below L3, resolution means paging a human — enter a card, complete a signup, click approve. Each interruption costs the full human response latency, and a job with several paywalls serializes those waits.
With a funded wallet and a spend policy (L3), the agent settles small charges itself — a metered API call, a dataset, a compute burst — and the loop never halts. Crypto rails suit this because a wallet is a keypair: no account creation, no card form, no human-shaped KYC funnel for a $0.03 API call. Payment-per-request protocols formalize the handshake; this page focuses on the economics of who gets interrupted.