The agentic commerce conversation has been stuck on the same question for months: will consumers let AI agents shop on their behalf?
That question is becoming obsolete.
This week’s signals—from payment orchestration platforms, stablecoin networks, merchant infrastructure providers and dispute resolution bodies—point to something more structural happening beneath the adoption headlines. The industry is quietly building a five-layer stack for autonomous purchasing, and the real competition is shifting to who controls each layer.
The Five-Layer Stack
The first layer is identity: proving which agent is acting. Ixopay, a payment orchestration platform, argued this week that payment tokens must carry agent identity and verifiable intent—not just a card credential. The question is no longer “is this a valid card?” but “is this a valid agent with permission to use this card for this purchase?”
The second layer is mandate: defining what the agent may buy. Stripe is deploying Shared Payment Tokens that integrate with Visa Intelligent Commerce and Mastercard Agent Pay network tokens, creating a unified primitive for agent checkout. The technical challenge is encoding user preferences—spend limits, merchant categories, delivery windows—into machine-readable authorization.
The third layer is payment rail: deciding how the transaction settles. Visa’s own data reveals the split. x402, the crypto-native protocol, averages $0.14 per transaction across 109.6 million transactions. Mastercard’s MPP averages $0.22 across 115,000 transactions. Both are micro-payment rails for machine-to-machine commerce. Traditional card rails still dominate consumer-facing transactions.
The fourth layer is merchant acceptance: getting sellers to trust agent-initiated orders. Here the gap is stark. Checkout.com reports 89% of merchants are preparing for agent commerce, 42% are testing it, but only 3% of transactions involve AI agents. Meanwhile, Product.ai’s April survey found just 14% of consumers trust AI to execute purchases without verification.
The fifth layer is dispute resolution: determining who pays when the agent gets it wrong. This may be the most unresolved layer. The American Arbitration Association warned this week that existing contract law—clickwrap, consent, ratification—doesn’t fit autonomous agents. Two competing models are emerging: Internet Court, a crypto-native system with 1,001 AI validators charging $0.85–$1.45 per case, and AAA’s own LCP framework, which uses machine-readable terms backed by traditional arbitration.
The Battle Lines
The five-layer stack is creating new competitive dynamics. Payment orchestration platforms like Ixopay want to be the merchant-side control layer. Card networks want to extend tokenization to agent identity. Stablecoin issuers argue their rails are native to machine-to-machine commerce. Commerce platforms like Salesforce are embedding agent workflows directly into checkout. And identity providers are positioning agent verification as the foundational layer.
Paymentology’s CTO Tim Joslyn told PYMNTS this month that 99% of existing issuer processing systems could already process an agentic payment. The payment rails aren’t the bottleneck. The real competitive battleground is trust infrastructure: identity, authorization, and fraud prevention.
Merchant Reality
For merchants, the calculus is complicated. Agent commerce promises lower conversion friction—no cart abandonment, no checkout hesitation, no comparison shopping fatigue. But it also introduces new liability ambiguity. Who handles a chargeback when the purchaser was an algorithm? Who verifies that the agent’s mandate matched the actual order?
Riskified’s June 30 global study found 73% of consumers are using AI in shopping and 70% are comfortable with agent purchases—up from 55% uncomfortable in Q1. But 32% cite payment security as their primary concern, and the liability gap for agent-initiated chargebacks remains unresolved.
The Week Ahead
The infrastructure fight is likely to intensify. Stripe’s SPT deployment is going live with Visa and Mastercard tokens. Salesforce’s Agentforce Commerce reached general availability in June. And the dispute resolution landscape is fragmenting between crypto-native and traditional-legal approaches.
The consumer trust gap—86% of shoppers verify AI recommendations before purchasing, 42% won’t trust AI for purchases over $25—suggests adoption will be gradual. But the infrastructure buildout is happening now, and the winners will be determined by who controls the permission layer.
The question is no longer whether agents can shop. It’s who proves they were allowed to.