Agentic Commerce Is Being Born Stillborn

I am painfully close to saying that agentic commerce is being born stillborn, not because the idea is bad, but because the industry is building it backwards.

And the idea is not bad. It might be one of the most important shifts the internet can have: software that can hold intent, move money, buy services, pay humans, settle obligations, and participate in the economy without a person manually clicking every button.

But look at how we’re actually approaching it, and something feels broken.

Everyone wants to be the standard

Every serious player is trying to build the standard for agentic payments, separately.

Coinbase and Cloudflare are pushing x402: a protocol built on HTTP 402, where services return payment instructions and clients pay programmatically, with no accounts, sessions, or API keys. Stripe and Tempo launched MPP, an open standard for agents and services to coordinate payments, microtransactions, recurring flows, the works. Google has AP2, centered on authorization, authenticity, and accountability. OKX has APP: agent payments, batch payments, pay-as-you-go, escrow-style settlement.

Individually, a lot of this work is genuinely impressive.

Collectively, it’s exhausting.

A category cannot mature when every serious player is racing to become the center of gravity before the market even knows what it needs. And the people who pay for that race are the builders, the ones working on SDKs, wallets, policy engines, observability, compliance, and real applications. Every protocol wants integrations. Every ecosystem wants examples. Every company wants “developer adoption.” So builders burn their time adapting to four competing surfaces instead of answering the only question that matters yet: does anyone actually want agents moving money at all?

Most of it is theatre

Which leads to the second problem. Most of agentic commerce today is still theatre.

The dominant demos are sub-dollar, sub-cent, testnet, simulated, or outright toy. Agents paying APIs. Agents buying data. Agents calling paid MCP tools. Agents paying other agents a few cents.

I respect these companies, and I respect the technical work, some of it is genuinely useful. But if the entire category presents itself as agents paying fractions of a cent to hit endpoints, we shouldn’t be surprised when serious customers don’t care.

Because the real question was never can an agent pay $0.01 for an API call?

The real question is trust

The real question is whether an agent can be trusted with economic authority.

Can it pay an invoice? Buy groceries? Reimburse an employee? Pay a contractor? File and pay taxes? Move money across borders? Hold a budget? Earn money? Invest under constraints? Subscribe, cancel, dispute, refund, escrow, audit, and explain?

Can it do any of this without borrowing your credit card like a glorified browser-automation script?

That is the real agentic commerce problem. Not payment. Trust.

I say this as someone who has spent the past year trying to build what I believed should exist: an agentic financial backbone. At first the question looked simple, how can agents hold money? It didn’t stay simple for long. The real question is: how can we trust agents to hold money?

Moving money is not just a technical action. It’s authority. Liability. Intent. Policy. Counterparty risk. Compliance. Auditability. Reversibility. It’s the ability to say, after something has happened, exactly why it happened, who authorized it, under which constraints, and what evidence existed at the time.

That’s the part almost no one wanted to ask seriously.

Why the familiar faces are winning

And maybe that’s why the early winners in this space aren’t necessarily the people with the sharpest product insight into agents. They’re the people who already had distribution, fintech networks, stablecoin relationships, compliance credibility, or prior proximity to companies like Stripe, Circle, Coinbase, Google, and the major payment networks.

That’s not evil. Finance has always worked this way.

But we should be honest about what’s happening. The market is not yet betting on agentic commerce as a bottom-up developer movement. It’s betting on familiar faces, existing rails, trusted institutions, and people who already had the right network before the category existed.

Understandable. But it’s not how a new industry becomes healthy. A healthy industry leaves room for the people closest to the problem, the builders who hit the walls early, who tried to make agents actually hold money, enforce policies, prevent bad actions, generate evidence, and survive contact with real financial workflows.

The control plane

Right now the category is still obsessed with rails and protocols. But protocols alone don’t create commerce. Commerce requires trust, and trust requires more than a signed transaction. It requires:

• policy before execution
• authority before signing
• evidence before settlement
• limits before autonomy
• auditability after action
• recourse when something goes wrong

This is why microtransactions are the wrong emotional center for agentic commerce. They’re useful, but they aren’t the main story. The main story isn’t “agents can pay websites.” It’s “agents can become economically useful without becoming financially dangerous.”

What real looks like

So the important use cases aren’t API access and paid data. They’re real-world workflows:

• paying your bills
• buying approved goods
• handling recurring vendor payments
• managing travel budgets
• negotiating and paying contractors
• moving funds across borders
• investing within strict mandates
• earning revenue and allocating it under policy
• operating inside a business with approvals, limits, receipts, and accountability

That’s when agentic commerce becomes real. Until then, most of what we call agentic commerce isn’t commerce, it’s monetized tool calling. And that is not a trillion-dollar category by itself.

The bar

The industry needs to stop pretending the hard part is getting an agent to pay.

The hard part is making sure it was allowed to pay. That the user meant it. That the merchant can trust it. That the payment can be audited, challenged, explained, constrained, and governed.

The hard part is not the transaction. It’s the control plane around the transaction. That’s the layer I still believe needs to exist.

If agentic commerce is going to become real, it won’t be because one more protocol wins a standards war. It will be because someone makes autonomous economic action safe enough, legible enough, and useful enough for the real world.

Not demos. Not testnet flows. Not one-cent endpoint access.

Real obligations. Real money. Real accountability.

That’s the bar. And right now, the industry is still far below it.