Agentic commerce and payments Report Interpretation

Overview

This industry commentary examines whether renewed AI-agent launches can revive agentic commerce and what that could mean for payments. Bernstein's conclusion is that adoption will take time, but the eventual need for trusted, tokenized and increasingly complex transactions makes the development more of a long-term opportunity than a disruption risk for payment networks and selected processors.

Core views

Bernstein says agentic commerce has re-entered the market discussion following launches including Meta's Muse and Instinct and Anthropic's Commerce Agents. Muse is described as a personal AI agent capable of executing multistep tasks across authorized services, while Anthropic's offering is an open-source framework for consumer shopping and merchant operations agents. The firm notes that earlier initiatives have not yet established a durable model: OpenAI abandoned Instant Checkout, and Google has focused more on infrastructure and standards than on controlling agentic payments. Although AI adoption and AI-originated traffic are growing rapidly, agentic commerce and AI-referred traffic still account for less than 1% of overall e-commerce activity today. The report argues that merchants continue to retain control of payments and checkout, which is constructive for incumbent payment participants. Current launches are generally focused on frictionless task execution and purchases rather than controlling the merchant payment flow; Muse, for example, can use single-use virtual cards. Bernstein attributes the difficulty of owning checkout to the operational demands of real-time pricing, conversion, catalog and inventory synchronization, loyalty and discounts, fulfillment, post-purchase processes and multi-basket transactions. Third-party agents face a chicken-and-egg problem: consumers need merchant participation before adoption can scale, while merchants may resist participating if they must surrender checkout or risk losing their direct customer relationship. This matters because the top 10 e-commerce merchants and marketplaces control about 50% of e-commerce. Bernstein expects the first successful uses to address complex commerce journeys, such as reservations and complicated bookings, rather than to replace retail shopping. The firm argues that discovery and information synthesis, not basic checkout, are the more meaningful consumer frictions. Consumers are increasingly using AI for discovery but are not yet broadly transacting through agents: a Visa survey cited in the report found that only 23% of consumers trust agents to make payments for them. AI-chatbot referrals represent under 1% of web traffic for the largest players, though the figure can be as high as 10% for certain direct-to-consumer brands; Bernstein's exhibits characterize generative-AI traffic as roughly 1-2% of total traffic. These data support its view that the sector remains in an iterative product-market-fit phase rather than being ready to take over a full shopping journey. The necessary infrastructure is nevertheless being built. Google launched the Universal Commerce Protocol in January 2026 as an open standard intended to connect agents and merchants across discovery, cart, checkout and post-purchase, and later expanded its effort into specialized councils for areas such as hotels and food delivery. Merchants and marketplaces are building their own agents, while Shopify is positioning itself as infrastructure through a catalog of more than 1 billion structured products and tools that let agents discover products, build carts and complete transactions while preserving merchant checkout, payments and fulfillment. Visa and Mastercard have also developed agent-focused tokens, standards and services. Bernstein stresses that autonomous agent-to-agent micropayments are much less developed than the attention around them suggests. Coinbase and Cloudflare's x402 initiative enables agents to discover resources, receive payment instructions through HTTP 402 responses, transact in stablecoins and access services without accounts, subscriptions or API keys. About one year after launch, volumes remain roughly $0.5-$1 million per month, average transaction size is around $0.20, and activity is concentrated in narrowly digital-native use cases such as pay-per-use APIs, datasets, web content and automation. The report concludes that these volumes remain several orders of magnitude away from material commercial relevance. For Visa and Mastercard, Bernstein directly rejects the view that agentic commerce is principally a disintermediation risk. It argues that a more digitized and complex transaction environment should increase transactions and value-added-services opportunities. The report views the card networks as trust layers rather than merely back-end money-movement systems, citing their dispute-management capabilities, standards, more than 7 billion credentials and over 100 million merchants. It expects all agentic transactions to be tokenized and points to Visa Intelligent Commerce and Mastercard Agent Pay as mechanisms that can add context and assurance to token credentials, including data on transaction type, token use and payer identity. Bernstein also sees opportunities in identity, verification, risk, readiness and AI billing; it notes that security solutions account for about 40% of Mastercard's value-added-services revenue and a teens percentage of Visa's value-added-services revenue. The report sees modern payment service providers as potential beneficiaries because merchants will need help navigating multiple protocols and products. Bernstein identifies Adyen and private Stripe as examples of PSPs positioned to manage this complexity. It highlights Adyen Agentic, intended to let enterprises sell through AI platforms without rebuilding their commerce systems for every channel, and Adyen's acquisition of Orb for AI usage-based billing. The outlook is less clear for digital wallets and card issuers: PayPal could extend its role around trust, protections, fraud, disputes, savings, rewards, personalization and payment optimization, while issuers could face more intense competition if AI assistants optimize card selection and rewards for each purchase.

Analysis framework

Bernstein begins with recent AI-agent launches and compares them with earlier agentic-commerce initiatives that have struggled to scale. It then evaluates adoption evidence, merchant integration constraints and protocol development before tracing the effects through payment networks, processors, wallets and issuers. The analysis uses transaction, traffic, trust and ecosystem data to distinguish long-term strategic potential from current commercial reality.

Methodology notes

• Competition & strategyValue chain analysis

  Commerce value-chain analysis across agents, merchants, marketplaces, payment networks, processors, wallets and issuers.

  The report assesses who controls checkout, who bears integration complexity and how more complex agent-led transactions could shift value toward trusted payment infrastructure and merchant-facing processors.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  Transmission from AI-agent adoption and merchant integration to payments participants.

  Bernstein links the development of agent protocols and merchant adoption to downstream effects on card tokenization, fraud prevention, payment services and competitive pressure on issuers.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Visa (V)

  Long-term beneficiary of more digitized, tokenized agentic transactions and associated trust, identity and risk services.

  Strengths

  Bernstein cites dispute management, standards, broad credential scale and agentic-token capabilities.

  Comparison

  Discussed alongside Mastercard as a payment network positioned to benefit rather than be disintermediated.

  Risks

  Near-term agentic-commerce adoption remains limited by consumer trust, merchant participation and infrastructure.

• Mastercard (MA)

  Long-term beneficiary of agentic commerce through payments, tokenization and value-added services.

  Strengths

  Bernstein highlights Agent Pay, trust infrastructure and security solutions representing about 40% of value-added-services revenue.

  Comparison

  Discussed alongside Visa as a network with a durable role in agentic payments.

  Risks

  Near-term commercial volumes and consumer adoption remain low.

• Adyen (ADYEY)

  Potential processor beneficiary as merchants need help connecting to multiple agentic-commerce protocols and channels.

  Strengths

  Adyen Agentic is designed to support sales through AI platforms without rebuilding commerce systems for each channel; the firm also acquired Orb for AI usage-based billing.

  Comparison

  Presented with private Stripe as a modern PSP that can help merchants manage integration complexity.

  Risks

  Merchant demand depends on the eventual scale and fragmentation of agentic commerce.

• PayPal (PYPL)

  A two-sided wallet that could expand its proposition as agents handle payments.

  Strengths

  Bernstein identifies potential roles in trust, protections, fraud and disputes, savings and rewards discovery, personalization and payment optimization.

  Weaknesses

  The report questions whether consumers will continue to value conventional guest-checkout friction in an agent-led environment.

  Comparison

  Contrasted with card issuers, which may face greater optimization-driven competition.

  Risks

  The ultimate consumer interface and wallet role remain uncertain.

Key data

• Agentic commerce / AI-referred e-commerce activityLess than 1% of overall e-commerce activityBernstein's current estimate, despite rapidly rising AI adoption and referral traffic.
• Consumer trust in agent payments23%Share of consumers trusting AI agents to execute payment transactions, according to a Visa survey cited by Bernstein.
• AI-chatbot referral trafficUnder 1% for the largest players; as high as 10% for certain DTC brandsThe report also characterizes generative-AI traffic as roughly 1-2% of total traffic.
• Top-10 merchant and marketplace concentration~50% of e-commerceTheir willingness to integrate agents is therefore pivotal to adoption.
• Agent-to-agent micropayment volume~$0.5-$1 million per monthRoughly one year after x402 launched; Bernstein characterizes commercial adoption as extremely limited.
• Average agent-to-agent micropayment~$0.20Reflects the current concentration in small digital-native use cases.
• Card-network scale7B+ credentials and 100M+ merchantsCited as support for Bernstein's view that cards retain a trust and acceptance advantage.
• Security value-added-services revenue mix~40% for Mastercard; teens% for VisaBernstein says these security businesses are growing well.

Impact & implications

Bernstein believes agentic commerce is unlikely to displace merchant checkout or card networks in the near term. If adoption broadens, more tokenized and information-rich transactions could expand network value-added services, while the need to manage fragmented protocols could strengthen the differentiation of modern merchant-facing PSPs. Wallets may gain new trust and optimization roles, whereas card issuers could face greater reward-optimization competition.

Risks

• Merchant and marketplace reluctance to relinquish checkout or customer relationships could slow agentic-commerce integration.
• Low consumer trust in authorizing AI agents to make payments could delay transaction adoption.
• Fragmented protocols, operational integration requirements and unresolved product-market fit may defer meaningful commercial scale.

What to watch

• Whether merchants and marketplaces integrate third-party agents while preserving checkout, payments and fulfillment control.
• Growth in AI-originated traffic and the conversion of AI discovery into completed transactions.
• Adoption of standards such as Google's Universal Commerce Protocol and payment-network agentic tokens.
• Whether AI billing, digital identity and cybersecurity services become meaningful additional payment-service opportunities.
• Whether agent-driven card and rewards optimization intensifies competition among card issuers.