Every major infrastructure category that has ever mattered has eventually bifurcated between open and closed models. The outcome of that bifurcation has shaped the economics of the category for decades. Cloud computing: the open source vs proprietary split created a world where AWS and Azure compete on the same open Linux foundation, rather than a world where each cloud provider's software only runs in their data centers. Payments: open banking frameworks vs closed card networks created the modern fintech ecosystem rather than a world where financial data is permanently locked in bank siloes. AI: open model weights vs closed API access is bifurcating right now in ways that will determine who controls the intelligence layer for the next decade.
Agentic finance is going through the same bifurcation now — earlier in its development than those categories when the architectural decisions were made, which means there's still time to build on the right side. The choice between open infrastructure and walled gardens will determine whether the agent economy is genuinely interoperable and competitive, or whether a small number of platforms control access to the financial rails that agents depend on.
What "walled garden" means in agent finance
The walled garden model for agent finance is custodial and vertically integrated. The provider holds your funds, operates the wallet infrastructure, enforces policies through their own systems, and routes payments through their own network. Everything is convenient and integrated — until it's not.
The garden walls become visible when you try to do something the provider didn't anticipate, or when your needs outgrow their roadmap, or when their pricing changes and you have no alternative. The walls become dangerous when the provider faces regulatory issues, experiences an operational incident, or changes their terms of service in ways that affect your agents' financial access. You're not building infrastructure — you're renting access to someone else's infrastructure on their terms, with revocable access and no portability.
For individual developers building proof-of-concept agents, walled garden infrastructure is genuinely convenient. The garden walls look like helpful features: everything works together, support is responsive, documentation is good. The costs of the walled garden model only manifest at scale, under operational stress, or when your needs diverge from the provider's roadmap.
What the open infrastructure model looks like
Open infrastructure for agent finance has three defining properties: non-custodial fund ownership, open protocol compatibility, and composability with the broader on-chain financial stack.
Non-custodial fund ownership means your agents hold their own keys. The infrastructure provider provides tooling, policy enforcement, and developer infrastructure — but they never control your funds. If they go offline, change their terms, or fail as a business, your agents' funds are intact and accessible. This is the property that transforms infrastructure from a service you rent to a platform you own.
Open protocol compatibility means your agent infrastructure works with any x402-compatible API, any wallet standard, any on-chain protocol. You're not restricted to the providers your infrastructure vendor has chosen to partner with. The network you can transact with is the open network — every agent on every provider that uses open standards — not just the walled garden's internal network.
Composability means your agents can interact with the full on-chain financial stack: lending protocols, yield sources, decentralized exchanges, other agent infrastructure providers. You can assemble the best options at each layer rather than accepting whatever the walled garden provides.
Why the choice compounds over time
The costs of walled garden infrastructure accumulate over time in specific, predictable ways. Early adopters often don't feel them because they're not yet at the scale where the structural problems manifest. But the trajectory is consistent:
- Transaction volumes grow. At low volume, per-transaction fees on closed platforms are ignorable. At enterprise scale — hundreds of thousands of transactions per month — the fee differential between competing on open infrastructure and being locked into a single provider becomes material. And in a closed garden, there's no competitive pressure to reduce those fees.
- Your agents need to transact outside the garden. As agent ecosystems mature, your agents will need to transact with counterparty agents on different platforms, access APIs from providers who haven't integrated with your walled garden, and interact with on-chain protocols the provider hasn't supported. Closed platforms don't interoperate. Every external interaction requires a custom integration or is simply unavailable.
- Enterprise customers apply scrutiny. When enterprise buyers ask "who controls the funds our agents operate?" and the answer is "our infrastructure provider," that conversation gets hard quickly. Non-custodial architecture answers this question correctly: the enterprise controls the funds. The infrastructure provider provides tooling. That's a clean separation that enterprise risk frameworks understand.
- Regulatory environments evolve. Custodial arrangements attract more regulatory scrutiny than non-custodial ones, because custodians are money transmitters. As regulatory frameworks for AI agent finance mature, custodial arrangements will face increasing compliance overhead that non-custodial providers simply don't have.
Proco's position — open infrastructure by design
Proco is built on the open infrastructure principle from the ground up. Non-custodial: your agents hold their own keys. Open protocols: x402-compatible out of the box, with AP2 support for complex agent-to-agent commerce. Composable: your agent wallets interact with the full on-chain stack, not just Proco's network. Auditable: every transaction is on-chain, independently verifiable, not in a proprietary ledger.
We compete on the quality of our tooling, the robustness of our policy engine, the developer experience, and the compliance coverage of our KYA layer. Not on lock-in. If a better tool exists for some layer of your stack, you should use it — our infrastructure is designed to compose with it, not to block it. That's the bet we're making: open infrastructure wins, the same way it has won in every comparable category. We're building the open standard infrastructure for agent finance.
Further reading: Why Non-Custodial Architecture Wins · How AI Uses Money — The Case for Open Standards