Agentic commerce: A practical guide to prepare your business

Digital commerce has been built around one assumption: a human is on the other side of the screen. We designed storefronts for visual appeal, optimised discovery for search engines, and built checkout flows for people to click, scroll, and type. Agentic commerce challenges that foundation.

As AI agents begin discovering products, comparing options, and completing purchases on behalf of consumers, merchants aren’t just building for people anymore. They’re building for agents acting with human intent. That shift is forcing merchants to rethink their technology stacks across product discovery, security, and payments. Merchants that treat agentic commerce as just another channel risk becoming invisible to agents, exposed to new forms of fraud, and overwhelmed by protocol complexity. Those that prepare intentionally can access new demand, reduce friction, and stay in control as the ecosystem evolves.

Discovery is becoming machine-first

Today, product discovery happens in places designed for humans – branded websites, marketplaces, ads, and search results. Trust and product discovery is built visually and emotionally through brand, design, reviews, and checkout experience. Agents don’t navigate your site and respond to design cues the way people do.

They query product feeds with constraints set by the user. Standard attributes like price, availability, delivery speed matter but also contextual ones from occasions, material sourcing, and brand values. That means the richness and structure of your product data (how well it answers real-world questions) matters just as much as the product itself.

This creates a subtle but important shift in how visibility is determined. Product feeds need to be able to serve standard attributes but also context-rich data that is interpretable by generative AI agents. Agents will also create a strain on product feed infrastructure with significantly more numerous, high-frequency real-time requests for inventory, pricing, and product details.

For merchants, this introduces a new kind of discoverability risk. You can have a strong brand and a beautiful storefront, but if agents can’t reliably parse your product data, you effectively don’t exist to the fastest-growing group of digital buyers.

Being agent-ready at the discovery layer means going beyond traditional SEO and thinking in terms of generative content consumption. That can start with relatively simple steps: auditing your product catalogue for structured, consistent metadata; enriching product descriptions with contextual attributes (use cases, constraints, delivery conditions); and exposing inventory and pricing through APIs and MCP endpoints or feeds that can support higher-frequency queries and updates.

This isn’t about replacing your storefront. It’s about making sure agents can see and understand you in the first place.

Trust becomes the new conversion layer

In traditional commerce, merchants have built their security against machines – risk systems were designed to spot and block scripted attacks and crawlers scraping their website data. Agentic commerce introduces a new category of traffic that can look very similar to malicious bot traffic. High-frequency inventory checks, automated checkout flows, and programmatic purchasing blur the line between legitimate automation and abuse.

Merchants today need to understand:

• Who the agent is (identity)

• What it’s allowed to do (mandate and permissions)

• What it intended to do vs what actually happened (intent vs execution)

Without these primitives, merchants are forced into a tradeoff: block automation to stay safe, or accept it and take on more risk. Liability becomes another layer of complexity. If an agent misinterprets intent, purchases the wrong quantity, or acts on outdated context, who is responsible?

Being agent-ready at the security layer means being able to verify that an agent is legitimate, enforce what it’s allowed to do, and maintain a clear audit trail of every action it takes. It also means being able to separate malicious bots from authorised agents, without shutting the door on automation entirely. 

There are practical steps merchants can begin taking here as well. Many are already revisiting bot mitigation strategies, not to block all automated traffic, but to differentiate and segment it. When it comes to trust, emerging protocols solve for agent identity, mandate, and intent validation. Agent-ready payment providers are inevitably integrating these protocols into their infrastructure and offering merchants the tools for agentic trust and security.

This trust layer becomes the new conversion layer. Without it, agentic commerce doesn’t scale safely.

Payment stacks need to keep up with many protocols

Checkout is where agentic commerce becomes real, and where complexity shows up quickly. Most payment flows rely on forms, redirects, CAPTCHAs, and UI patterns that agents don’t navigate naturally. Meanwhile, the ecosystem is evolving quickly and unevenly. Different AI platforms and networks are defining different protocols for discovery, authentication, and payment execution. Some follow structured tokenized flows, and others mimic human behaviour. In practice, this means agentic demand won't come from a single source. It will come from multiple LLMs, agent types, and protocols, often at the same time. 

For merchants, this creates complexity reminiscent of when digital marketing and ads first exploded – many new platforms, standards, and flows to orchestrate simultaneously. Supporting one protocol isn’t enough. Traffic can come from multiple LLMs, multiple agentic frameworks, and continuously evolving protocols and network standards.

Being agent-ready at checkout means your payment infrastructure can accept agent-initiated transactions across protocols, opening up demand from any agentic surface. It means handling tokenized credentials, mandate and intent data, and agent authentication in a way that’s secure, compliant, and adaptable. And it turn, this exposes what functions your checkout supports on machine-readable endpoints.

Merchants don’t need to solve this fragmentation themselves, but they do need to plan for it. In the near term, that may mean working with providers that can orchestrate different payment flows and support emerging standards, rather than building tightly around a single platform or maintaining direct support for several protocols. The merchants that win will be the ones that build for interoperability from the start.

Post-checkout becomes part of the trust loop

Agentic commerce doesn’t end at payment. In many ways, post-checkout is where trust is proven. Disputes, refunds, and chargebacks will increase if agents act outside of user expectations or operate on stale context. Merchants will need better ways to establish that a transaction was properly authorised, mandates were respected, and agent behaviour stayed within defined limits.

Post-checkout workflows become part of the control plane for agentic commerce. Being agent-ready means your systems can surface agent authorisation as compelling evidence, support automated dispute handling (alt: support automated resolution flows), and adapt policies to a world where “buyer error” may actually be “agent misinterpretation". 

In practice, this is where more automated resolution flows will start to emerge. Merchants can begin preparing by ensuring agentic transaction data is captured with enough context to support disputes, and by evaluating how their current refund and dispute processes would hold up in a world of non-human buyers. Without this layer, merchants risk absorbing the cost of liability.

Agent-ready is a system, not a feature 

When you step back, a pattern emerges. Agent-readiness isn’t about a single integration or protocol. It’s about building a system that works across the entire lifecycle of agent-initiated commerce:

• Discovery that works for machines, not just people

• Trust infrastructure that separates legitimate agents from malicious bots

• Payments that operate across evolving protocols

• Post-checkout workflows that protect merchants from new liability risks

Treating any one of these in isolation creates blind spots: Discovery without trust invites abuse. Payments without protocol flexibility create lock-in. Trust without post-checkout protection shifts liability back to the merchant. The merchants that succeed in agentic commerce will be the ones that think holistically.

How Airwallex is approaching agent-readiness

Agentic commerce is still early, and the standards are still forming. That’s exactly why infrastructure matters. Airwallex’s approach is to build universal rails for agentic commerce: modular, protocol-agnostic infrastructure that helps merchants accept agent-initiated demand and payments across platforms without being tied to a single ecosystem.

At the payments and trust layer, this means building capabilities merchants will need most urgently: infrastructure to handle tokenized credentials and mandates, payment rails that support emerging agentic protocols, and risk systems that can verify agent identity, analyse behaviour, and help merchants control liability exposure.

That said, agentic commerce doesn't stop at payments. Discovery and product visibility are becoming just as critical, and this is where merchants have to make their biggest design decisions. Ensuring that product data is structured and contextual is one side of the coin, making sure the infrastructure serving it is reliably accessible to agents is another. It requires coordination between merchant systems, platforms, and partners.

Our goal isn’t to replace merchant experiences. It’s to abstract the complexity of infrastructure, agentic interfaces, and protocol compliance so merchants can focus on their products, remain trusted, and stay in control as commerce becomes increasingly autonomous.

The merchants who prepare now won’t be playing catch-up later

Agentic commerce won’t replace human shopping overnight. But it doesn’t need to. In the US, nearly half of consumers (47%) have already used AI for at least one shopping-related task. This data signals the beginning of an agentic AI era where shoppers rely on intelligent tools not just to browse, but to make decisions too. As agents move from assisting to action, even partial adoption (especially among high-frequency, high-value shoppers) will reshape how demand flows to merchants. The real risk isn’t that every customer switches to agents tomorrow. It’s what your most valuable customers do, and your systems aren’t ready to meet them there.

Being agent-ready today is about preparing for a future where discovery, trust, and payments are no longer purely human-only systems. The merchants that build for that future early won’t just keep up with the shift; they’ll help define it.