a16z: Where Are the “Blue Ocean” Startup Opportunities in Agent Payments?
a16z: Where Are the “Blue Ocean” Startup Opportunities in Agent Payments?
AI Agents are quickly evolving from helpful copilots into autonomous “digital operators” that can research, decide, and execute tasks end-to-end. When the actor that initiates a transaction is no longer always a human, payments stop looking like a classic “click → checkout → pay” retail flow—and start looking more like machine-native commerce.
In Sam Broner’s a16z crypto essay, agents “will pay like locals, not tourists”: the winning agents won’t behave like one-off shoppers, but like repeat business operators with negotiated terms, credit relationships, and high-frequency settlement needs. (See the original: Tourists in the bazaar: Why agents will need B2B payments — and why stablecoins will get there first.)
This shift creates a new payments frontier for crypto founders: agent payments infrastructure—spanning stablecoins, account abstraction, identity, compliance, dispute handling, and credit.
1) Why “Agent Payments” Don’t Fit the Old Internet Checkout Model
The last internet era optimized for a single moment: a human deciding to pay. In the agent era, the system must support something different:
- High-frequency, low-value transactions (API calls, compute, data subscriptions)
- Programmatic purchasing (agents executing policies, budgets, and constraints)
- Bundled workflows (travel booking + insurance + local transport + reservations, approved as one intent)
- Cross-border vendor graphs (agents routing to global providers and dynamically swapping suppliers)
Broner’s key insight is that successful agents will act like “locals”: they repeatedly transact with preferred counterparties, earn better terms, and rely on credit/invoicing more than consumer card rails. That’s structurally closer to B2B commerce than to retail checkout. (a16z crypto)
2) Why Stablecoins Are Becoming the Default “Settlement Layer” for Agents
Agentic commerce needs money that is:
- Always-on (nights/weekends included)
- Programmable (policy-driven spending, conditional release, streaming)
- Global by design (instant settlement across borders)
- Composable (works with smart contracts, escrow, automation)
That maps naturally to stablecoins and onchain settlement rails—especially as mainstream institutions increasingly treat stablecoins as serious infrastructure. For example, Visa announced USDC settlement availability for U.S. institutions on December 16, 2025, explicitly framing stablecoins as a modern settlement layer for payments networks. (Visa press release)
At the same time, major policy and research bodies have been formalizing how stablecoins fit into the global financial system, including the IMF’s late-2025 deep dive on stablecoin market structure, use cases, and risk tradeoffs. (IMF: Understanding Stablecoins)
Founder takeaway: stablecoin payments aren’t just a consumer “pay with crypto” button anymore—they’re increasingly a B2B settlement primitive that agents can use for metered services, cross-border procurement, and automated treasury.
3) The Wallet Layer Is Changing: From “Human Wallets” to “Agent-Controlled Accounts”
If agents can initiate payments, wallets need new capabilities:
- Fine-grained permissions (what can be bought, from whom, and under what conditions)
- Session keys / scoped keys (short-lived authority, limited blast radius)
- Budgeting and spend controls (per vendor, per category, per time window)
- Safe delegation and revocation (users can pause, dispute, or rotate credentials)
Ethereum’s account model is also moving in this direction. The Pectra upgrade activated on Ethereum mainnet on May 7, 2025, bringing EIP-7702 and enabling EOAs to temporarily delegate execution to smart-contract logic—pushing “smart account” UX forward without forcing everyone to migrate immediately. (Ethereum Foundation: Pectra Mainnet Announcement, plus an ecosystem explainer: Alchemy: What is ERC-4337?)
Founder takeaway: agent payments will likely converge on programmable wallets and policy-based authorization, not just basic key signing.
4) The “Blue Ocean” Startup Opportunities in Agent Payments
Below are the most promising build directions implied by the “locals, not tourists” framework—translated into concrete crypto startup surfaces.
A. Agent-to-vendor B2B rails (invoicing, net terms, reconciliation)
Agents will need more than instant pay: they’ll need invoices, credits, and accounting-grade records.
Opportunities:
- Stablecoin invoicing + automated reconciliation (ERP-friendly exports, onchain proofs)
- Net-7 / net-30 terms with onchain settlement guarantees
- Payment orchestration for agent platforms (one user approval → many vendor payouts)
Why crypto helps:
- Immutable payment logs
- Programmable settlement
- Cross-border liquidity without correspondent banking complexity
B. Credit and working capital for agents (underwriting + risk engines)
If agents pay like locals, credit becomes a core primitive.
Opportunities:
- Underwriting for agent platforms based on cashflow, vendor SLAs, chargeback history
- Stablecoin credit lines with onchain monitoring (covenants, automated margin calls)
- Factoring for agent-generated receivables
Hard part (and moat):
- Fraud and loss control when “the buyer” can be duplicated infinitely
C. Escrow, arbitration, and dispute systems designed for autonomous buyers
Broner highlights that dispute resolution is a feature cards have historically done well—but agent commerce will demand new models, especially for digital goods and metered services. (a16z crypto)
Opportunities:
- Programmable escrow with usage-based release (pay-per-call, pay-per-token, pay-per-minute)
- Machine-readable receipts and proof-of-delivery standards
- Arbitration markets specialized by vertical (compute, data, ad inventory, travel)
D. Micropayments and streaming payments for machine-native services
Agents will purchase things humans rarely pay for directly: tiny increments of value.
Opportunities:
- Streaming stablecoin payments for compute and APIs
- Sub-cent pricing abstractions (aggregation, batching, probabilistic settlement)
- Intent-based “one approval, continuous spend” models with safety rails
E. Identity, provenance, and “proof of authority” for agents
When an agent pays, vendors must know: Is this agent authentic? Is it allowed to do this? This is fertile ground for crypto-native identity.
Opportunities:
- Verifiable credentials for agent identity (org-bound, role-bound, policy-bound)
- Attestation layers (who deployed this agent, what code is running, what constraints exist)
- Privacy-preserving compliance proofs (selective disclosure for KYC/AML where required)
F. Compliance tooling for stablecoin flows in agent networks
As stablecoin usage scales, compliance expectations scale with it—especially for platforms coordinating many vendors and geographies.
Opportunities:
- Onchain compliance routing (rule engines, jurisdiction filters, sanctions screening integrations)
- Audit trails that connect: user intent → agent execution → settlement outcome
- Treasury controls for businesses paying hundreds of vendors per day
5) What Founders Should Build First: A Practical Wedge
Agent payments can feel huge, so here are three pragmatic wedges that align with real demand:
- A stablecoin invoicing + reconciliation API for agent platforms (start with “make finance teams happy”).
- A programmable authorization layer (policies, budgets, revocation, scoped keys) that can plug into smart accounts.
- A dispute-ready receipt standard for agent purchases (machine verifiable, vendor signed, easy to audit).
If you can make “agent spend” legible—who approved what, why, and what was delivered—you’ll unlock trust, credit, and scale.
6) Where OneKey Fits: Securing the Human “Root of Trust” in Agent Commerce
Even if agents execute day-to-day transactions, users and businesses still need a high-security root layer for:
- Setting/rotating agent permissions
- Approving high-risk or high-value intents
- Managing long-term keys and recovery
A hardware wallet can serve as that root of trust by keeping private keys offline and making critical approvals explicit on a secure device. In an agent economy, this matters because the failure mode isn’t just “phishing a human”—it’s silently delegating authority to an agent that can spend at machine speed.
Conclusion
Agents won’t pay like tourists clicking through checkout pages. They’ll pay like locals: repeat counterparties, negotiated terms, and credit-driven B2B workflows—exactly where stablecoins and programmable onchain settlement shine. (a16z crypto)
For builders, the blue ocean isn’t “another pay button.” It’s the missing infrastructure that makes agent commerce safe and scalable: policy-based wallets, stablecoin settlement, identity and attestations, invoicing, arbitration, and credit—all designed for autonomous actors operating at global scale.
