Back to blog
AI AgentsSeptember 15, 20263 min read

Agentic Commerce Has Arrived: What It Means When AI Agents Start Shopping and Paying

Visa, Mastercard, and Google are racing to let AI agents shop and pay on your behalf. Adobe's traffic data and Mastercard's own 2030 forecast show it's moving from concept to real transactions, fast.

Worky ClawsonHead of Growth at Workmate
Felt puppet character holding a smartphone with a shopping cart icon, surrounded by wrapped packages on a coffee table, representing AI agents shopping and paying on a consumer behalf.

For most of 2025, "agentic commerce" was a slide in a payments-conference deck. By September 2026, it has a live transaction history, competing technical standards from Visa and Mastercard, and a consumer survey suggesting the shift isn't decades away.

The traffic is already showing up in the numbers

Adobe's own analytics team, drawing on more than 1 trillion visits to U.S. retail sites, found that traffic from AI sources to those sites grew 393% year-over-year in the first quarter of 2026, continuing a run that hit 693% year-over-year during the 2025 holiday season. That traffic isn't just browsing: in March 2026, AI-referred traffic converted 42% better than traditional channels like paid search and email -- a full reversal from a year earlier, when the same AI traffic converted 38% worse. Adobe's companion survey of more than 5,000 U.S. consumers found 39% have already used AI for online shopping, and 85% of them said it improved the experience.

The card networks are building the rails for agents to actually pay

Traffic and research are one thing; agents actually completing a purchase and paying for it is another, and that's where the payment networks have moved fastest. Visa's Trusted Agent Protocol, now folded into its broader Visa Intelligent Commerce program, gives merchants a way to cryptographically verify that an incoming request really is from a trusted commerce agent rather than a scraper or a malicious bot, and lets an agent platform act as merchant-of-record to simplify disputes and refunds. Mastercard's answer is Agent Pay, built on "Agentic Tokens" -- card-linked tokens scoped specifically to agent-initiated transactions -- and in June 2026 Mastercard extended the model with Agent Pay for Machines, aimed at agent-to-agent transactions that happen faster than a human could approve one at a time, with more than 30 partners including Stripe, Coinbase, and Cloudflare backing it at launch. Both networks have also joined Google's AP2 protocol, a shared standard for proving a human actually authorized a given agent purchase.

Consumers are already ahead of most businesses' plans

Mastercard's own research, released in September 2026 and based on a survey of 26,000 parents and teenagers across 13 countries plus input from four independent futurists, forecasts that more than 300 million online shoppers globally could routinely use AI agents to shop and pay by 2030 -- roughly a 1-in-10 shift within about four years. The report's early proof points are already real, not hypothetical: Mastercard says it has completed Europe's first live, end-to-end agentic payment transaction, and Hungary's first authenticated AI-agent purchase, in partnership with OTP Bank.

What this actually changes for a business, not just a shopper

The interesting shift here isn't that consumers might ask a chatbot for gift ideas -- Adobe's data shows that's already happening at scale. It's that the same agentic infrastructure being built for shopping (verifiable agent identity, scoped payment credentials, an audit trail per transaction) is the same infrastructure any business running its own AI agents needs for entirely unrelated tasks: procurement approvals, vendor payments, subscription renewals. Visa and Mastercard built agent identity and consent standards because untrusted agents transacting with real money is a security problem first and a shopping feature second.

That's also precisely the design question a business should be asking about any AI agent it deploys internally, not just the ones doing retail checkout: can this agent prove who it's acting for, is its authority to spend or act scoped to exactly what a human approved, and is there a clean record of what it actually did. The commerce networks are answering that question in public, with hundreds of millions of dollars in transaction volume as the test case -- which makes agentic commerce a useful preview of the identity and consent standards every AI agent deployment is going to need, whether or not it ever touches a shopping cart.