The architecture for autonomous machine payments is consolidating around a two-layer split that has been implicit for months but is now formally acknowledged. Below sits a protocol layer where agents discover services, negotiate prices, and establish identity — x402, the Agent Payments Protocol (AP2), and the Machine Payments Protocol occupy this space. Above sits the settlement layer, where stablecoins actually move between agent wallets. The distinction sounds academic, but it determines which projects compete with each other and which are complementary infrastructure.
The Protocol Layer: Negotiation and Identity
Three protocol standards are vying for the below-the-settlement layer. Coinbase’s x402 standard, which repurposes HTTP’s 402 Payment Required status code, handles price negotiation and payment within the HTTP request-response cycle. The Agent Payments Protocol (AP2) covers similar ground with a different design philosophy, while the Machine Payments Protocol addresses the narrower question of machine-to-machine settlement instruction formatting. Separately, the Model Context Protocol — an open standard initially designed for giving AI models access to external tools — is being adapted for financial data access. N3XT’s launch of an MCP implementation for live banking data shows the standard extending beyond pure model context into payment-adjacent territory.
The Settlement Layer: Stablecoins as the Default Rail
Above the protocol layer, stablecoins have effectively won the settlement role for agent payments. USDC on Base dominates the existing integrations, largely because Coinbase’s x402 implementation defaults to it and because Base’s low fees and fast finality suit microtransactions. USDT on Tron and Solana remains the settlement rail for a large share of existing off-chain stablecoin flows, though its presence in agent-specific infrastructure is thinner. The settlement layer’s job is narrow: move value atomically, cheaply, and with enough finality that the protocol layer above can treat the transaction as complete. The protocol layer does not care which stablecoin settles the payment, only that it settles.
What the Split Means for Competition
The layering creates a competitive landscape that does not map onto traditional payment industry boundaries. Projects building protocol-layer standards — x402, AP2, MCP implementations — are competing with each other for adoption among agent developers and API providers. Projects building settlement infrastructure — stablecoin issuers, wallet providers, chain teams — are competing on cost, speed, and reliability, not on agent-facing features. A company like Skyfire, which builds agent-specific wallet infrastructure, straddles both layers, which is either an advantage or a strategic distraction depending on execution. The risk for any single project is that winning the protocol layer yields limited pricing power if settlement remains commoditized — the same problem payment networks have always faced.
Open Questions in the Stack
Several structural questions remain unresolved. Interoperability between x402 and AP2 is not yet specified, meaning an agent using one protocol cannot seamlessly pay a service using the other — a fragmentation risk that mirrors early HTTP versus Gopher dynamics. The MCP standard’s expansion into financial data access raises questions about whether it will absorb protocol-layer payment functionality or remain a complementary standard. And the settlement layer’s reliance on a small number of stablecoins creates concentration risk: if USDC’s issuer freezes an agent’s wallet, the protocol layer has no fallback. These are not hypothetical concerns. They are the concrete engineering decisions that will determine whether agent commerce scales beyond demo transactions into production volume.