In a recent CNBC interview, Yannick Malling — CEO of Public.com — described something that would have sounded like science fiction two years ago. Everyday investors on Public's platform can now deploy AI agents that monitor oil prices and execute protective put options, scan for covered call opportunities, move cash into bond portfolios based on yield conditions, and act on natural language intent: "Increase my position if valuations compress another 15%."
The numbers are real. Fifty percent of AI conversations on Public.com lead to a transaction within 24 hours. The agents are deterministic once a strategy is set — no hallucinations, no random execution. Malling calls this the fourth era of investing: phone brokers gave way to discount web brokers, which gave way to mobile neo-brokers, which are now giving way to agentic brokerages.
He's right about the eras. He's right that this is real. And the market signal matters: if half of AI-assisted financial conversations result in a transaction within a day, the agentic finance category is not emerging — it has already arrived.
The structural problem no one is talking about
Public.com's numbers prove demand. But every one of their agents operates through a human brokerage account. The agent has no financial identity of its own. It cannot hold funds natively, cannot settle with another agent, and cannot enforce spending policies at the wallet layer. It's AI layered on top of human financial infrastructure — sophisticated software sitting on top of rails that were designed for a different actor entirely.
That model works for their consumer use case. An individual investor has a brokerage account. The agent acts on their behalf. The human remains the legal and financial principal. This is a legitimate and valuable product. But it is architecturally bounded.
The moment you move outside that consumer framing, the model breaks. And the cases where it breaks are precisely the cases that define agentic finance at scale:
- An agent with its own treasury. Not a human account as the underlying principal with the agent as a delegate — an agent that holds its own working capital, denominated in programmable assets, without requiring a human in the ownership chain.
- Agent-to-agent settlement. Two agents, no human intermediary, settling a commercial transaction. Credit card tokens require a human cardholder. Regulated bank models are custodial. Neither handles this case natively.
- Programmable spending policies enforced at the wallet layer. Not application logic sitting on top — constraints baked into the financial object itself. An agent that literally cannot execute a transaction outside defined parameters, regardless of what the application layer instructs it to do.
- Sub-cent, high-frequency transactions. Agent micropayments break human-scale AML frameworks, pricing models, and fee structures. Human banking infrastructure was not designed to process ten thousand $0.003 transactions per hour.
This is an infrastructure problem, not a UX problem
It's tempting to frame the gap as a product design challenge — better interfaces, smarter agents, more automation layered on existing rails. That framing is wrong. The limitation is structural, not presentational.
Human financial infrastructure was built on a set of assumptions: a legal person is the financial actor, that person has identity documents, transactions occur at human cadence, and AML/KYC frameworks operate against that person. Every piece of regulation, every compliance framework, every API contract, every fee structure was designed around those assumptions.
Agents violate all of them simultaneously. They are not legal persons. They operate at machine speed. They may have no fixed domicile. They may be ephemeral — spun up for a single transaction and destroyed. The identity primitives, the compliance models, and the transaction economics of human finance don't transfer.
The x402 protocol — backed by Google, Anthropic, Visa, AWS, and Cloudflare, with over 35 million transactions — is building the protocol layer that addresses this. But protocol is not enough on its own. Protocol defines how agents communicate about value. You still need a wallet layer: financial identity for agents, treasury management, programmable policy enforcement, and settlement infrastructure built natively for non-human actors.
What Public.com actually proved — and what it left open
Public.com validated the demand side of the market. Investors want agents that act on their behalf. Fifty percent conversion from conversation to transaction is extraordinary. It proves that people are ready to delegate financial decision-making to AI, provided the agent presents options and the human retains final approval.
What it did not validate — and what it structurally cannot validate with its current architecture — is the supply side: infrastructure that can support agents as first-class financial actors. That validation has to come from a different layer of the stack.
Public.com's agents are impressive. But they will eventually hit the ceiling that human rails impose. The agents that break through that ceiling won't be running on brokerage accounts. They'll be running on financial infrastructure built for them — wallets with their own identity, treasuries with programmable policy, settlement that doesn't require a human principal.
Where the infrastructure gap needs to be filled
The gap between "AI acting on a human's account" and "AI as a first-class financial actor" requires new primitives at every layer:
- Financial identity for agents. Not a user account with agent permissions — a financial identity native to the agent, portable across platforms, with its own compliance profile.
- Non-custodial treasury. Working capital the agent controls without a custodian holding it on behalf of a human principal. The agent is the principal.
- Policy enforcement at the wallet layer. Spending limits, counterparty restrictions, and transaction type controls that are enforced by the financial object, not by application code that can be bypassed.
- Settlement designed for machine economics. Sub-cent transactions, high throughput, programmable settlement finality — none of which fit into human-scale banking fee structures.
This is the layer Proco is building. Not another interface on top of human rails. Financial infrastructure that treats agents as the financial actor, not as a sophisticated delegate to a human account.
The timing is not a coincidence
Public.com's announcement didn't happen in isolation. It's one of several signals in the same week pointing at the same thing. The agentic brokerage is real. Agent-driven commerce at scale is real. The demand is validated. The market is moving.
What isn't moving fast enough is the infrastructure underneath it. The protocol layer is being built. The wallet layer is being built. The compliance frameworks for non-human financial actors are in early formation. The question is whether the infrastructure catches up to the demand before the ceiling becomes visible to everyone — or whether an entire generation of agentic finance products gets built on architectural constraints that will eventually require painful rewrites.
The agentic finance stack needs its own rails. Not because the existing rails are bad. Because agents aren't humans.
Further reading: Building on x402 — The Open Payment Protocol for AI Agents · From Automation to Autonomy — Why AI Agents Need Their Own Money · Why Legacy Financial Rails Break Under Agent Load