This white paper argues that AI agents will need payment infrastructure designed for machine-speed commerce, not just repurposed human payment rails. Its core claim is that current options both fall short: centralized systems are easy to freeze or monitor, while transparent blockchains expose transaction data that can reveal strategy, counterparties, and competitive positioning.
The paper’s thesis is that agent payments require four things at once: privacy, permissionless access, neutrality, and a way for agents to earn capital from computation. From that view, it presents Zcash as the strongest current fit because it combines Proof of Work with shielded transactions, while arguing that Proof of Stake systems are structurally weaker for censorship resistance and neutrality in an autonomous agent economy.
@joshs @aquietinvestor @zooko @ZcashFoundation
WP2 Why AI Agents NeedPrivate-Payment Rails v1.pdf (63.9 KB)
1 Like
The mining bootstrap in Section 7 is the part that connects to what we’re building.
Nordic Shield runs a hosted mining program on Zcash with an attestation protocol (NSM1) that commits the full participant lifecycle to the chain - entry, ownership, deployment, hosting payments, renewals, transfers, exits.
Every state change goes into a BLAKE2b Merkle tree anchored on mainnet via shielded memos. Three anchors confirmed since March 27.
The relevance to your thesis: if an agent mines ZEC to bootstrap its own capital (your zero-capital bootstrap loop), it still needs a way to prove what it owns and what it’s done without revealing that to competitors. That’s the attestation problem. The mining earns the capital. The shielded attestation layer makes the capital and its provenance verifiable to the owner while staying invisible to everyone else.
The developer tooling gap you flag in Section 13 is real. We have a verification SDK on crates.io
( crates.io: Rust Package Registry ) with a WASM build, so an agent running in a browser or Node environment can verify Merkle proofs without a server round-trip. That’s one piece of the agent integration surface. The API layer (26 endpoints, https://pay.frontiercompute.io/protocol/info) is another.
The throughput math in Section 12 is honest and I appreciate that. At current TPS, batched settlement is the realistic model. Our anchor automation already works that way - accumulate events, anchor periodically. An agent settlement layer would follow the same pattern.
1 Like
Thank you for your response and your reading through the white paper. Looking forward to working with you more on this specifically.
The thesis is coming from a real world event with Alibaba. I’m sure you’ve been tracking it since last year. Here is Alibabas real world agent case study.
https://arxiv.org/pdf/2512.24873
Good link. Capability side is moving fast. What it doesn’t solve is the part your paper gets right - agents can execute but can’t settle privately or prove history without exposing it.
Zcash has the privacy primitive. The missing piece is attestation on top. Commit state changes to shielded memos, anchor periodically, prove specific claims without revealing the full record. That’s what memo-field protocols do.
The 80-tx-per-block reality from your Section 12 is fine for this. Agents don’t need per-action settlement. They need periodic anchoring with selective disclosure between anchors. That works today on Zcash at current throughput.
1 Like