How AI Uses Money: The Case for Open Standards in Agent Commerce

This post summarizes our Web Summit 2025 presentation: "How AI Uses Money" with links for further research.

In 1995, e-commerce was ready to explode, but it was stalled. We had browsers and websites, but no trust. The creation of SSL/TLS—an open, cryptographic trust layer—wasn't just an update; it unlocked a trillion-dollar economy.

Today, we are at the exact same moment with AI. AI agents are ready to transact... but they're stalled. We are missing the trust standards for AI commerce.

The Problem: Fragmentation Before We Begin

Today, merchants face an impossible choice: allow "bots" to transact and risk fraud liability, or block them entirely and lose revenue. Without interoperable standards for agent identity and authorization, the default is to block.

Meanwhile, legacy payment infrastructure wasn't designed for machines. Agents need instant settlement, global connectivity, and programmable guardrails, but they're stuck with systems built for human speed, business hours, and manual error handling.

Instead of collaborating on shared solutions, companies and teams are currently building competing solutions, or forking and fragmenting nascent proposed standard solutions. Multiple new protocols launched in 2025 (AP2, ACP, x402, ATXP, KYAPay, etc.). AI providers, fintech platforms, and new payment startups are creating proprietary systems that do not interoperate.

What's Missing

For humans, we agreed upon trust and payment standards long ago:

  • Identity: KYC/KYB regulatory standards and practices (with government and network rules)
  • Authorization: OAuth 2.0 and related extensions for delegated access
  • Policy: Card networks and banks enforce limits, authentication (e.g., 3-DS), and fraud controls
  • Payments: Card/bank rails (PCI/EMV/ACH/RTP), secured over TLS/HTTPS during transport
  • Audit: Comprehensive reporting and accountability (network + PSP + merchant logs)

For agents? Nothing. Every field is blank. Different companies are reinventing different wheels.

History Shows Open Standards Work

Open standards unlock trillion-dollar economies:

  • TLS secured the web
  • TCP/IP connected the internet
  • SMTP made email universal
  • SMS made every phone interoperable
  • Shipping containers revolutionized global trade

These succeeded because everyone — including market competitors — benefited from interoperability. Open compounds value. Closed caps it.

What the Ecosystem Needs

Four foundational components must work together:

1. Agent Identity (KYA: Know Your Agent)

Universal cryptographic proof that works across all protocols. Agents need verifiable credentials that establish identity, delegation, and authorization.

2. Authorization, Policy & Safety

Verifiable delegation chains, programmable guardrails, and safety rules that travel with the transaction. When an agent acts, we need to know: who operates it, under what constraints, and what it's allowed to do.

3. Transparency & Accountability

Immutable audit trails. When something goes wrong, liability must be clear. Who authorized what, when, and why? This isn't optional, it's table stakes for any system handling real money.

4. Payment Protocol Interoperability

We need standardized protocols that let agents pay for resources, pay humans, get paid, and transact with each other, regardless of the underlying payment infrastructure.

All of this must be built on open standards. No single owner, no walled gardens.

What Catena Labs Is Building

Catena Labs is partnering with AI providers, agent developers, fintechs, stablecoin issuers, and banks to define open trust standards for the agent economy. We’re building an AI-native financial institution on interoperable protocols so agent commerce is open, auditable, and safe at internet scale.

Join the Conversation

The window for building open standards is narrow. Early architectural choices compound—we've seen this with the internet, email, and mobile networks. The decisions we make in the next 12-18 months will determine whether we build interoperable protocols or fragment into incompatible silos.

Want to collaborate?

Email: hello@catenalabs.com Twitter: @catenalabs LinkedIn: Catena Labs

References: Who’s Building What

Core Agentic Protocols (Comms, Tools, Commerce)

  • Agent-to-Agent (A2A) Protocol — Google's proposal for agent discovery and communication.
  • Model Context Protocol (MCP) — Anthropic-originated de facto standard for connecting agents to tools, data, and services.
  • Cloudflare Agents SDK — SDK + edge runtime for deploying durable agents in Cloudflare.
  • CrewAI — Multi-agent orchestration framework (Python).
  • LangChain — Agent framework (LangGraph, LangSmith) for building production LLM apps.
  • LlamaIndex — Agent/RAG framework with document/workflow tooling.
  • Vercel AI SDK — TypeScript toolkit for apps/agents on Node/Next.js.

Agent-Facing Payment & Commerce Protocols

Network / Provider Initiatives

Durable Orchestration & Workflows

  • Inngest — Durable, event-driven workflows and queues for agents.
  • Trigger.dev — Long-running jobs, retries, and observability for AI agents.
  • n8n — Visual workflows and AI nodes with 400+ integrations.
  • Zapier AI Actions — Expose 6k+ app actions to agents.

Foundational Identity & Authorization Standards