How the right architecture is the key to enabling agentic commerce in B2B

The first thing to accept is that "the buyer" is no longer just a person using a storefront. Buyers are now made up of a diverse group of human and non-human clients, including autonomous AI agents, automated systems, and traditional procurement platforms, each with very different traffic patterns and trust profiles. These include: 

• The human specialist: Still uses a portal, but is increasingly assisted by an AI co-pilot or internal seller-side agents.

• The buyer agent: Acts autonomously, turning complex tasks (like "reorder with split delivery, respect cap") into separate, machine-speed API calls, using the Model Context Protocol (MCP) to securely discover and carry out commerce functions.

• The procurement system: Connects via punchout or other standardized integrations, which demands faster and smarter reasoning capabilities.

• The IoT device: Generates replenishment orders based on telemetry events from a production line, requiring zero human intervention.

• The partner marketplace: Forwards authenticated orders from an unknown end-customer, demanding a reliable, authenticated API path.

Each of these client types needs the exact same underlying capability: addressable, contract-aware commerce building blocks. None of them benefit from a standard web-based commerce storefront. All of them break the assumptions baked into legacy ecommerce systems.

Serving this mix of non-human clients comes down to two simple principles:

1. Total reachability

Every piece of relevant commerce logic must be reachable without going through a rendered webpage. Things like contract-aware pricing (tiered, negotiated, customer-specific), location-aware inventory, and delivery dates to a specific destination must be visible via APIs.

2. Context and security

Before a non-human client acts, it needs to know what the customer is allowed to buy, including corporate hierarchies, approval limits, and contracted assortments. An agent that cannot see the business rules will either break them or refuse to act. Furthermore, this interaction is only safe with an identity framework built for non-human callers, featuring scoped credentials and a full audit trail per agent or device.

This is the point where MACH (Microservices, API-first, Cloud-native, Headless) transitions from an industry buzzword into a concrete architectural requirement. While often discussed as a general philosophy for modernizing IT, each of these four pillars maps directly to a specific, functional requirement of agentic commerce:

• Microservices handle non-human traffic. An agent might issue forty pricing calls for every one order. A fleet of agents might constantly ping inventory status. Microservices allow you to scale your pricing or inventory services independently of checkout, apply rate limits to agent traffic, and ensure a runaway agent doesn't crash your human channel/storefront.

• API first creates an equal playing field. Every commerce capability, including pricing, cart, checkout, customer, and inventory is defined as an API contract before any UI exists. There’s no privileged frontend path and backend path. Instead, there’s one path and the storefront is just one of its clients. An MCP server, an agent protocol endpoint, a procurement system, or an IoT replenishment service is another, on equal footing.

• Cloud native eliminates the need for constant upgrades. A continuously delivered platform lowers TCO in a way that finance teams notice because the line item that used to be a major version upgrade every two or three years simply disappears. More importantly, when the platform ships native support for a new agent protocol or a new identity model for non-human clients, you can adopt it instantly.

• Headless decouples logic from the screen. Headless architecture ensures the storefront is treated as just another client rather than the core of the system. When an internal sales assistant, an autonomous buyer agent, or a telemetry-driven IoT device needs to transact, each interacts with the same underlying business logic. These non-human callers authenticate and consume the exact same commerce building blocks as the web portal. This removes the need to maintain legacy partner APIs.

One of the clearer examples of an API first, MACH-aligned platform in this category is Commercetools. Every capability is exposed as a documented API. The data model accommodates B2B constructs such as business units, associates, and quotes.

The SaaS delivery model removes the upgrade tax. And its MCP capabilities give agents a structured, permissioned way to discover and invoke commerce functions, which is exactly the integration point a serious agent strategy needs. The right choice depends on your existing architectural landscape, internal technical maturity, and channel mix. Vincit can help you make the best decision. 

Agentic commerce in B2B is still evolving. While internal seller-side agents for quote automation and service resolution are already operational at some manufacturers and distributors, the landscape is maturing in stages. The more complex middle ground, where agents negotiate against contract terms autonomously or handle various exceptions, is currently being defined. For now, the most effective approach maintains a human-in-the-loop for final commitments while granting agents broader autonomy for the retrieval and orchestration of data.

This means that while the ultimate vision is evolving, the foundational architectural requirements are already clear. Investing in the future means ensuring your core commerce logic is exposed as a service that can be consumed instantly by any non-human system. These platform decisions are not theoretical, they determine if you’re ready to adopt the next wave of agentic capabilities.

Key takeaways

• The real question is whether your commerce architecture is actually capable of handling autonomous clients.
• The buyer is no longer necessarily a person. Commerce platforms designed only to render web pages for humans will struggle to survive the next decade of B2B traffic. 
• Composable architecture is a business strategy, not an aesthetic preference. It’s the only approach that allows agentic commerce, IoT replenishment, and direct procurement to coexist. 
  
  

Are you ready for agentic commerce?

Whether you’re weighing a move to a composable platform like Commercetools, planning a targeted decomposition of your legacy platform, or looking for a practical first agent use case to run against your existing stack, Vincit has solved these exact architectural problems for retailers, manufacturers, and distributors across the Nordics and beyond.

Get in touch with us to set up a working session with our experts.

Written by our Commerce Solution Architects, Jarno Räisänen and Sami Asmundi.