One day soon, you might log into your payments dashboard after a strong campaign week — to realize that something has changed.Â
Traffic is up, but a growing share of new customers are no longer arriving through search or ads. Instead, they’re coming through AI agents that browse, recommend, and even complete purchases on their behalf.
That shift is the foundation of agentic commerce. AI is moving beyond product discovery into real transaction execution, and payment leaders need to understand what that means for the architecture underneath.
In our latest webinar, Architecting Agentic Payments, Andrew Sjogren, Director of Product Marketing, IXOPAY, is joined by Jill Willard, CTO, IXOPAY, Marco Conte Founder, Congrify, and Dwayne Gefferie, Founder, Gefferie, to break down how agentic commerce is reshaping the payment stack. The message is clear: this is a fast-evolving layer you need to prepare for now.
Agentic Commerce Is Moving Fast
Agentic commerce is moving faster than most payment teams expected. Jill Willard pointed to early signals already reshaping the funnel, with more shopping journeys starting inside large language models instead of traditional channels.Â
Industry projections are only adding fuel.Â
Mastercard estimates that agent-led commerce could represent almost fifty percent of transactions by 2030. And Stripe has already introduced new capabilities, including support for emerging agentic commerce protocols like PX402.
Right now, most use cases still sit in the recommendation phase. AI helps customers discover products, compare options, and narrow decisions, while a human still approves the final checkout. But the market is clearly pushing toward execution.
For you, the real question is no longer whether AI agents will transact. Rather, it’s how quickly that shift will hit your risk controls, routing strategy, and your payment stack as a whole.
How Agentic Checkout Adds New Layers to Your Stack
Agentic commerce will not rip out your existing payments infrastructure. As Jill Willard explained, the change happens above it. AI agents introduce new layers that sit at the top and edges of your stack, where intent and policy must be handled before a transaction even reaches authorization.
That means you should be building for:
Intent management: capturing what the agent is trying to do (book, reorder, subscribe) before payment execution
Policy enforcement: deciding what an agent is allowed to complete without human approval
Monitoring and escalation: flagging when an agent’s behavior shifts from normal shopping to risky automation
Jill noted that “intent now becomes a new piece of the payment stack.” Today, merchants and PSPs are already carrying that burden, whether they are ready or not.
The New Risks Merchants Need to Prepare For
As Marco Conte warned, agent-led checkout will change dispute and fraud patterns faster than most payment teams are ready for. The issue is not volume. It is that the buyer behavior behind each transaction becomes harder to interpret.
For example, today a legitimate customer might browse for ten minutes, log in from their usual device, and complete checkout normally. Your fraud and dispute systems understand that pattern.
But with an agent, the same purchase could happen instantly through an automated session or virtual machine, with no browsing trail and different metadata. If that customer later disputes the charge, it becomes much harder to tell whether it was fraud, an agent mistake, or a valid delegated purchase.
You should expect three immediate risk shifts:
Chargebacks will look different
When a customer delegates the purchase, disputes will rise around “I didn’t authorize this” or “the agent did it.” Liability is still likely to land on you (the merchant), not the issuer.Product misalignment and hallucinations become real refund drivers
Marco gave the simplest example: a customer asks an agent for red shoes, and it buys blue ones instead. That becomes a dispute, not a UX issue.Fraud signals get murkier
As Dwayne Gefferie noted, there will no longer be a clean device fingerprint. Transactions may originate from virtual machines, bots, or automated sessions that deviate from the customer’s usual pattern.
If you do not redesign monitoring and dispute workflows now, agent traffic might expose those gaps quickly.
The Protocol Landscape
Right now, there is no unified protocol winning agentic commerce. As Jill explained, these standards form a capability ladder, and you should choose based on how much execution and risk you are ready to absorb.
UCP / ACP → discovery + assist, with humans still approving checkout
APW → wallet execution layer, bridging recommendation into payment flow
Visa Trusted Agent Protocol → bot identity, trust, and safer delegation
PX402 → highly programmatic automation, crypto-heavy, and higher risk
The key point: these protocols can work together.
If you are a merchant seeing early traffic from Gemini or ChatGPT, start with discovery-first rails. If you are ready to test execution, ladder up carefully with strong intent capture, monitoring, and dispute controls before scaling automation.
Phase Zero: Build the Foundations Before Agents Start Transacting
If you want to prepare for agentic commerce, Phase Zero starts with intent capture.Â
Jill stressed that intent must be traceable and replayable, so you can always answer why a transaction was executed, by which agent, and under what authority.
Your next step is tokenization with intent metadata. Best practice is storing intent alongside the payment token, but keeping it PSP-agnostic, since many providers cannot support metadata storage inside their own tokens.
This also forces a shift in data strategy. As Dwayne Gefferie noted, intent introduces unstructured data that you must safely store and control. You need visibility across both structured payment fields and agent behavior context.
Finally, monitoring is non-negotiable. Marco warned that as agent traffic scales from zero-point-something percent to ten percent, you must track fraud shifts, authorization impacts, and dispute spikes in real time.
The Panel’s Playbook for Getting Started
The panel’s advice was practical. You should not yet bet the whole house on agentic commerce, but standing still is the wrong move. Start with focused experimentation now so you’re not scrambling to catch up later.
Key actions for payment leaders:
Run proofs of concept in 2026 so you understand where AI agent flows fit into your stack, before volume arrives
Match rollout to customer readiness, since most shoppers still want a human approving the final purchase today
Track AI-driven discovery traffic immediately, because searches starting in Gemini, ChatGPT, or Perplexity will turn into payment transactions over time
Stay involved in protocol and partner conversations early, even if you’re not building the standards yourself
The takeaway: be strategic, start small, and build learning loops before agent payments begin to scale.
Agentic Payments Add a New Channel Layer
Agentic commerce is the beginning of a new payments channel sitting above your existing stack. You’re not rewriting your authorization rails or ripping out your PSP layer. The shift happens above it, where agents introduce new requirements for intent capture and policy enforcement.
To succeed, you need the right foundations in place:
Tokenization that supports delegated transactions
Intent traceability so every agent purchase is auditable end-to-end
Protocol flexibility as standards evolve in parallel
Intelligence and risk controls to monitor fraud, disputes, and authorization shifts
Do you want the full discussion and technical breakdown? Watch the IXOPAY webinar.