Google's AP2 and the x402 work Coinbase announced with Google give AI agents a way to pay, which moves them from suggesting financial actions to executing them: an agent with a mandate can transact, execute a strategy, and move capital on a user's behalf as a participant in the economic system rather than an assistant beside it.
Protocol note (2026): This post is a snapshot from September 2025. AP2 is an authorization layer, not a stablecoin-payment spec by itself. x402 is a settlement primitive, not an implementation of AP2. OpenAI's commerce work is usually discussed as ACP (Agentic Commerce Protocol), separate from Google's UCP (Universal Commerce Protocol).
What agentic finance requires
Agentic finance is the onchain economy where AI agents execute financial actions with real assets, real liquidity, and verifiable outcomes. It depends on three capabilities:
- Autonomous reasoning: agents plan, adapt, and decide without a human approving every step.
- Non-custodial execution under constraints: user intent becomes action without giving up safety or transparency.
- Unified liquidity access: agents operate across fragmented DeFi ecosystems and chains without losing composability.
Protocols that define agent payments
Three launches set the early standards:
-
Google AP2: a protocol for AI agents to transact using stablecoins with safety controls and cryptographic verification.
-
x402 (Google + Coinbase): Coinbase's stablecoin rails combined with Google's AI infrastructure, connecting web2 and web3 for agent payments.
-
Agentic Commerce (OpenAI): a framework for agent-driven commerce with agent-to-agent transactions, negotiation, and settlement over standardized payment protocols.
Together they describe how an agent receives delegated intent, how its execution is verified, and how the user keeps cryptographic control over what it may spend.
These two videos show agentic payments working end to end, with mandates, agent negotiation, settlement, and feedback loops that are easier to follow in motion than in a diagram:
Why stablecoins are the settlement layer
Stablecoins combine the programmability of crypto with the price stability commerce needs, so an automated system can move value and still know what it moved. A volatile asset would make every agent decision a market bet as well.
Web3 infrastructure supplies the rest of what these protocols need:
- Trust-minimized execution: agents act with cryptographic guarantees rather than opaque permissions.
- Composable primitives: standards like AP2 can integrate with any blockchain that supports the required verification methods.
- Permissionless deployment: new agent capabilities ship without a central gatekeeper approving them.
- Programmable constraints: smart contracts enforce the boundaries an agent can operate within.
Payments come first, and agent-led trading comes after them, as mandates and session keys evolve into trade instructions that let an agent execute swaps, manage risk, and settle positions inside boundaries you define. That is where I want to take the specialized agent interfaces I built on Masterbots, from commerce toward capital operations.
From payments to trading
This post builds on my earlier article on AI and Web3, where I looked at how agents and decentralized infrastructure could merge into new product experiences.
The standards are new, stablecoins are the most practical rails available, and teams are already shipping on them. The open engineering work is mandate design, cross-chain liquidity routing, auditability, and guardrails.
I would start with delegated agent payments on open standards like AP2, then add trading, portfolio management, and capital movement under the same mandate model, with each new capability confined by explicit constraints. The chains, networks, and marketplaces underneath will keep changing, so the part worth designing carefully now is the mandate: what a user grants an agent wallet-level authority to do, and how that grant is revoked.