Article
Agentic Foundations: Getting enterprise systems ready for AI commerce
The infrastructure changes you need to support AI assistants and help your business grow.
Our Agentic Foundations blog series breaks down the infrastructure barriers standing between today’s agentic demos and production-scale commerce. It details what it will take to move from experimentation toward real-world solutions.
We’ve explored the inventory gap and the trust paradox. But solving data and trust is only part of the equation. Autonomous commerce also depends on whether enterprise systems can support machine-led execution at scale.
The next constraint sits deeper in the stack: legacy systems were never designed for this. Even though early agentic commerce demos have been convincing, architectural gaps surface quickly once autonomous flows hit real enterprise stacks.
Fragmented systems can't keep up with AI tools
Most enterprise retailers have spent decades assembling their architecture: a product information management (PIM) system for product data, a dedicated pricing engine, loyalty systems, an order management system (OMS) for fulfillment, and a payment gateway to tie it all together. And on top of this, no two stacks look the same. What works for a fashion retailer running Salesforce Commerce Cloud with Manhattan OMS may look entirely different from a sporting goods retailer operating on a homegrown commerce platform with custom inventory and fulfillment logic.
These systems were not designed to communicate in real time. They were built around sequential handoffs at discrete checkpoints. Think of a relay-race model: the commerce platform hands off to the tax engine, which hands off to the payment processor. A human shopper navigating a checkout flow provides the sufficient time for these systems to catch up.
Agentic commerce disrupts this traditional flow. Agents don’t wait for the next page to load. They query pricing, inventory, and configuration repeatedly and simultaneously throughout a single interaction.
Why custom integrations won't work
When faced with shifting from human-led to agentic commerce, merchants typically see two paths — neither particularly sustainable.
The first is a rip-and-replace approach. This requires rebuilding the entire stack from scratch to become agent-native. For most enterprise retailers, that is commercially unrealistic. Years of customiation, integrations, and business logic are deeply embedded across legacy systems.
The second is custom integration where every internal system — PIM, OMS, tax, payments, fulfillment — needs direct connection into every emerging agent protocol, from OpenAI’s ACP to Google’s UCP. But that quickly introduces significant integration complexity, where every new platform launch becomes a bespoke engineering project.
As we noted in our recent paper, patching systems for today’s demos without preparing for autonomous flows is building on sand. It works until faced with volume arrives; or a new platform launches.
What your platform needs to handle AI tools
Agentic commerce places very different demands on enterprise systems than traditional human-led checkout flows. Sessions need continuous recalculation of pricing, taxes, and inventory availability before a transaction is ever created, yet most systems were not designed to operate at the speed and frequency autonomous agents require.
Inventory management also becomes more complex: stock must be reserved and released dynamically throughout a live session, as the timing windows built for human checkouts are often too slow and imprecise for autonomous flows.
In delegated commerce models — where an agent acts on behalf of a shopper who may not be actively present — transaction construction and execution can happen at entirely different moments in time. That separation introduces a fundamentally different operating model from the simultaneous, user-driven flows most enterprise systems were originally built to support.
The infrastructure solution
Just as Adyen unified the fragmented world of global payments into a single integration, our ambition is to build neutral, scalable infrastructure, and solve complexity for merchants in the next era of commerce.
The infrastructure solution will be a layer that sits between agent developers and the merchant’s existing stack. This will function as a translator, handling the high-frequency queries and stateful sessions the agents demand, while interacting with the merchant’s legacy systems of record at a pace they can handle.
Integrating once into an abstraction layer will allow merchants to gain architectural reversibility. This means they remain in control regardless of which agent protocol wins or which new consumer surface emerges. Essentially, the complexity remains outside the core stack and in the infrastructure where it belongs.
Agentic commerce will arrive gradually, then all at once. The merchants best positioned to capture the uplift are those building these foundations today — not by chasing every new demo, but by rearchitecting how they execute.
Explore more insights on the infrastructure shaping agentic commerce in Adyen’s Agentic Commerce hub.